Feedback control of dynamic systems [6th ed., International ed] 9780136019695, 0136019692, 9780135001509, 0135001501

Citation preview

Feedback Control of Dynamic Systems Sixth Edition

Gene F. Franklin Stanford University

J. David Powell Stanford University

Abbas Emami-Naeini SC Solutions, Inc.

Upper Saddle River Boston Columbus San Francisco New York Indianapolis London Toronto Sydney Singapore Tokyo Montreal Dubai Madrid Hong Kong Mexico City Munich Paris Amsterdam Cape Town

Vice President and Editorial Director, ECS: Marcia J. Horton Acquisition Editor: Andrew Gilfillan Editorial Assistant: William Opaluch Director of Team-Based Project Management: Vince O’Brien Marketing Manager: Tim Galligan Marketing Assistant: Mack Patterson Permissions Project Manager: Vicki Menanteaux Senior Managing Editor: Scott Disanno Production Project Manager: Clare Romeo Senior Operations Specialist: Alan Fischer

Operations Specialist: Lisa McDowell Art Director: Kenny Beck Cover Designer: Kristine Carney Manager, Rights and Permissions: Zina Arabia Manager, Visual Research: Beth Brenzel Image Permission Coordinator: Debbie Latronica Manager, Cover Photo Permissions: Karan Sanatar Composition: TexTech International Full-Service Project Management: TexTech International Printer/Binder: Courier Westford Typeface: 10/12 Times

Copyright © 2010, 1999, 1989 by Pearson Higher Education, Inc., Upper Saddle River, NJ 07458. All rights reserved. Manufactured in the United States of America. This publication is protected by Copyright and permissions should be obtained from the publisher prior to any prohibited reproduction, storage in a retrieval system, or transmission in any form or by any means, electronic, mechanical, photocopying, recording, or likewise. To obtain permission(s) to use materials from this work, please submit a written request to Pearson Higher Education, Permissions Department, One Lake Street, Upper Saddle River, NJ 07458. The author and publisher of this book have used their best efforts in preparing this book. These efforts include the development, research, and testing of theories and programs to determine their effectiveness. The author and publisher make no warranty of any kind, expressed or implied, with regard to these programs or the documentation contained in this book. The author and publisher shall not be liable in any event for incidental or consequential damages with, or arising out of, the furnishing, performance, or use of these programs. Pearson Education Ltd., London Pearson Education Singapore, Pte. Ltd Pearson Education Canada, Inc. Pearson Education—Japan Pearson Education Australia PTY, Limited Pearson Education North Asia, Ltd., Hong Kong Pearson Educación de Mexico, S.A. de C.V. Pearson Education Malaysia, Pte. Ltd. Pearson Education, Inc., Upper Saddle River, New Jersey Library of Congress Cataloging-in-Publication Data Franklin, Gene F. Feedback control of dynamic systems / Gene Franklin, J. David Powell, Abbas Emami-Naeini. — 6th ed. p. cm. Includes bibliographical references and index. ISBN-13: 978-0-13-601969-5 ISBN-10: 0-13-601969-2 1. Feedback control systems. I. Powell, J. David, 1938– II. Emami-Naeini, Abbas. III. Title. TJ216.F723 2009 2009025662 629.8
3—dc22

10 9 8 7 6 5 4 3 2 1 ISBN-13:978-0-13-601969-5 ISBN-10: 0-13-601969-2

To Gertrude, David, Carole, Valerie, Daisy, Annika, Davenport, Malahat, Sheila, and Nima

This page intentionally left blank

Contents Preface

1

An Overview and Brief History of Feedback Control 1.1 1.2 1.3 1.4

2

A Perspective on Feedback Control Chapter Overview 2 A Simple Feedback System 3 A First Analysis of Feedback 6 A Brief History 9 An Overview of the Book 14 Summary 16 Review Questions 16 Problems 17

Dynamic Models 2.1

2.2 2.3
2.4

2.5

3

xiii

1

20

A Perspective on Dynamic Models 20 Chapter Overview 21 Dynamics of Mechanical Systems 21 2.1.1 Translational Motion 21 2.1.2 Rotational Motion 27 2.1.3 Combined Rotation and Translation 36 2.1.4 Distributed Parameter Systems 38 2.1.5 Summary: Developing Equations of Motion for Rigid Bodies 40 Models of Electric Circuits 41 Models of Electromechanical Systems 45 Heat and Fluid-Flow Models 50 2.4.1 Heat Flow 50 2.4.2 Incompressible Fluid Flow 54 Historical Perspective 60 Summary 62 Review Questions 63 Problems 64

Dynamic Response 3.1

1

74

A Perspective on System Response 74 Chapter Overview 75 Review of Laplace Transforms 75 3.1.1 Response by Convolution 75 3.1.2 Transfer Functions and Frequency Response 3.1.3 The L− Laplace Transform 87 3.1.4 Properties of Laplace Transforms 89

80

v

vi Contents

3.2

3.3 3.4

3.5 3.6


3.7
3.8

3.9

4

3.1.5 Inverse Laplace Transform by Partial-Fraction Expansion 3.1.6 The Final Value Theorem 93 3.1.7 Using Laplace Transforms to Solve Problems 94 3.1.8 Poles and Zeros 96 3.1.9 Linear System Analysis Using MATLAB 97 System Modeling Diagrams 102 3.2.1 The Block Diagram 102 3.2.2 Block Diagram Reduction Using MATLAB 107 Effect of Pole Locations 108 Time-Domain Specifications 116 3.4.1 Rise Time 116 3.4.2 Overshoot and Peak Time 117 3.4.3 Settling Time 118 Effects of Zeros and Additional Poles 120 Stability 130 3.6.1 Bounded Input–Bounded Output Stability 130 3.6.2 Stability of LTI Systems 131 3.6.3 Routh’s Stability Criterion 132 Obtaining Models from Experimental Data 140 3.7.1 Models from Transient-Response Data 142 3.7.2 Models from Other Data 146 Amplitude and Time Scaling 147 3.8.1 Amplitude Scaling 147 3.8.2 Time Scaling 148 Historical Perspective 149 Summary 150 Review Questions 151 Problems 152

A First Analysis of Feedback 4.1

4.2

4.3

4.4

91

170

A Perspective on the Analysis of Feedback 170 Chapter Overview 171 The Basic Equations of Control 171 4.1.1 Stability 173 4.1.2 Tracking 174 4.1.3 Regulation 174 4.1.4 Sensitivity 175 Control of Steady-State Error to Polynomial Inputs: System Type 4.2.1 System Type for Tracking 179 4.2.2 System Type for Regulation and Disturbance Rejection The Three-Term Controller: PID Control 186 4.3.1 Proportional Control (P) 187 4.3.2 Proportional Plus Integral Control (PI) 187 4.3.3 PID Control 188 4.3.4 Ziegler–Nichols Tuning of the PID Controller 192 Introduction to Digital Control 198

178 183

Contents 4.5

5

6

vii

Historical Perspective 203 Summary 205 Review Questions 206 Problems 207

The Root-Locus Design Method

220

A Perspective on the Root-Locus Design Method 220 Chapter Overview 221 5.1 Root Locus of a Basic Feedback System 221 5.2 Guidelines for Determining a Root Locus 226 5.2.1 Rules for Plotting a Positive (180◦ ) Root Locus 228 5.2.2 Summary of the Rules for Determining a Root Locus 233 5.2.3 Selecting the Parameter Value 234 5.3 Selected Illustrative Root Loci 236 5.4 Design Using Dynamic Compensation 248 5.4.1 Design Using Lead Compensation 249 5.4.2 Design Using Lag Compensation 254 5.4.3 Design Using Notch Compensation 255 5.4.4 Analog and Digital Implementations 257 5.5 A Design Example Using the Root Locus 260 5.6 Extensions of the Root-Locus Method 266 5.6.1 Rules for Plotting a Negative (0◦ ) Root Locus 266
5.6.2 Consideration of Two Parameters 270
5.6.3 Time Delay 272 5.7 Historical Perspective 274 Summary 276 Review Questions 278 Problems 278

The Frequency-Response Design Method

6.1

6.2 6.3

6.4 6.5 6.6 6.7

A Perspective on the Frequency-Response Design Method 296 Chapter Overview 297 Frequency Response 297 6.1.1 Bode Plot Techniques 304 6.1.2 Steady-State Errors 315 Neutral Stability 317 The Nyquist Stability Criterion 319 6.3.1 The Argument Principle 320 6.3.2 Application to Control Design 321 Stability Margins 334 Bode’s Gain–Phase Relationship 341 Closed-Loop Frequency Response 346 Compensation 347 6.7.1 PD Compensation 348 6.7.2 Lead Compensation 348

296

viii Contents 6.7.3 6.7.4 6.7.5 6.7.6
6.7.7
6.7.8

PI Compensation 360 Lag Compensation 360 PID Compensation 365 Design Considerations 371 Specifications in Terms of the Sensitivity Function Limitations on Design in Terms of the Sensitivity Function 377
6.8 Time Delay 381
6.9 Alternative Presentation of Data 382 6.9.1 Nichols Chart 382 6.10 Historical Perspective 386 Summary 386 Review Questions 388 Problems 389

7

State-Space Design 7.1 7.2 7.3 7.4

7.5

7.6

7.7

7.8 7.9

7.10

7.11


373

413

A Perspective on State-Space Design 413 Chapter Overview 413 Advantages of State-Space 414 System Description in State-Space 416 Block Diagrams and State-Space 421 7.3.1 Time and Amplitude Scaling in State-Space 424 Analysis of the State Equations 425 7.4.1 Block Diagrams and Canonical Forms 425 7.4.2 Dynamic Response from the State Equations 436 Control-Law Design for Full-State Feedback 442 7.5.1 Finding the Control Law 443 7.5.2 Introducing the Reference Input with Full-State Feedback 451 Selection of Pole Locations for Good Design 455 7.6.1 Dominant Second-Order Poles 456 7.6.2 Symmetric Root Locus (SRL) 457 7.6.3 Comments on the Methods 466 Estimator Design 466 7.7.1 Full-Order Estimators 466 7.7.2 Reduced-Order Estimators 472 7.7.3 Estimator Pole Selection 476 Compensator Design: Combined Control Law and Estimator 478 Introduction of the Reference Input with the Estimator 491 7.9.1 A General Structure for the Reference Input 492 7.9.2 Selecting the Gain 501 Integral Control and Robust Tracking 502 7.10.1 Integral Control 503 7.10.2 Robust Tracking Control: The Error-Space Approach 505 7.10.3 The Extended Estimator 516 Loop Transfer Recovery (LTR) 519

Contents
7.12 Direct Design with Rational Transfer Functions 524
7.13 Design for Systems with Pure Time Delay 527 7.14 Historical Perspective 530 Summary 533 Review Questions 534 Problems 536

8

Digital Control 8.1 8.2

8.3

8.4

8.5


8.6

8.7

9

558

A Perspective on Digital Control 558 Chapter Overview 559 Digitization 559 Dynamic Analysis of Discrete Systems 561 8.2.1 z-Transform 561 8.2.2 z-Transform Inversion 562 8.2.3 Relationship between s and z 565 8.2.4 Final Value Theorem 566 Design Using Discrete Equivalents 568 8.3.1 Matched Pole–Zero (MPZ) Method 571 8.3.2 Modified Matched Pole–Zero (MMPZ) Method 575 8.3.3 Comparison of Digital Approximation Methods 575 8.3.4 Applicability Limits of the Discrete Equivalent Design Method 576 Hardware Characteristics 577 8.4.1 Analog-to-Digital (A/D) Converters 577 8.4.2 Digital-to-Analog (D/A) Converters 578 8.4.3 Anti-Alias Prefilters 578 8.4.4 The Computer 579 Sample-Rate Selection 580 8.5.1 Tracking Effectiveness 581 8.5.2 Disturbance Rejection 581 8.5.3 Effect of Anti-Alias Prefilter 582 8.5.4 Asynchronous Sampling 583 Discrete Design 583 8.6.1 Analysis Tools 583 8.6.2 Feedback Properties 585 8.6.3 Discrete Design Example 586 8.6.4 Discrete Analysis of Designs 588 Historical Perspective 590 Summary 591 Review Questions 592 Problems 593

Nonlinear Systems

599

Perspective on Nonlinear Systems Chapter Overview 600

599

ix

x Contents 9.1 9.2

9.3 9.4
9.5

9.6

10

Introduction and Motivation: Why Study Nonlinear Systems? 600 Analysis by Linearization 602 9.2.1 Linearization by Small-Signal Analysis 603 9.2.2 Linearization by Nonlinear Feedback 608 9.2.3 Linearization by Inverse Nonlinearity 608 Equivalent Gain Analysis Using the Root Locus 609 9.3.1 Integrator Antiwindup 615 Equivalent Gain Analysis Using Frequency Response: Describing Functions 619 9.4.1 Stability Analysis Using Describing Functions 625 Analysis and Design Based on Stability 629 9.5.1 The Phase Plane 630 9.5.2 Lyapunov Stability Analysis 636 9.5.3 The Circle Criterion 642 Historical Perspective 648 Summary 649 Review Questions 650 Problems 650

Control System Design: Principles and Case Studies 10.1 10.2 10.3

10.4 10.5 10.6 10.7 10.8

660

A Perspective on Design Principles 660 Chapter Overview 661 An Outline of Control Systems Design 662 Design of a Satellite’s Attitude Control 667 Lateral and Longitudinal Control of a Boeing 747 684 10.3.1 Yaw Damper 689 10.3.2 Altitude-Hold Autopilot 696 Control of the Fuel–Air Ratio in an Automotive Engine 702 Control of the Read/Write Head Assembly of a Hard Disk 709 Control of RTP Systems in Semiconductor Wafer Manufacturing Chemotaxis or How E. Coli Swims Away from Trouble 731 Historical Perspective 739 Summary 741 Review Questions 742 Problems 743

717

Contents

Appendix A

Laplace Transforms A.1

757

The L− Laplace Transform 757 A.1.1 Properties of Laplace Transforms 757 A.1.2 Inverse Laplace Transform by Partial-Fraction Expansion A.1.3 The Initial Value Theorem 769 A.1.4 Final Value Theorem 770

Appendix B

Solutions to the Review Questions

Appendix C

MATLAB Commands Bibliography Index

801

793

xi

788

772

766

This page intentionally left blank

Preface In this Sixth Edition we again present a text in support of a first course in control and have retained the best features of our earlier editions. For this edition, we have substantially rewritten Chapter 4 on the Basic Properties of Feedback, placing the material in a more logical order and presenting it in a much more effective (bottom up?) manner. We have also updated the text throughout on how computer-aided design is utilized to more fully reflect how design is carried out today. At the same time, we also strive to equip control system engineers with a basic understanding so that computer results can be guided and verified. In support of this updating, MATLAB
referrals have been updated and include some of the latest capabilities in that software. The case studies in Chapter 10 have been retained and a new case study of the emerging Bioengineering field has been added. A Historical Perspective section has been added at the end of each chapter in order to add to the knowledge of how all these concepts came into being. Finally, in order to guide the reader in finding specific topics, we have expanded the Table of Contents to include subsections. The basic structure of the book is unchanged and we continue to combine analysis with design using the three approaches of the root locus, frequency response, and state-variable equations. The text continues to include many carefully worked out examples to illustrate the material. As before, we provide a set of review questions at the end of each chapter with answers in the back of the book to assist the students in verifying that they have learned the material. In the three central chapters on design methods we continue to expect the students to learn how to perform the very basic calculations by hand and make a rough sketch of a root locus or Bode plot as a sanity check on the computer results and as an aid to design. However, we introduce the use of MATLAB early on in recognition of the universal use of software tools in control analysis and design. Furthermore, in recognition of the fact that increasingly controllers are implemented in imbedded computers, we again introduce digital control in Chapter 4 and in a number of cases compare the responses of feedback systems using analog controllers with those having a digital “equivalent” controller. As before, we have prepared a collection of all the MATLAB files (both “m” files and SIMULINK
files) used to produce the figures in the book. These are available at the following Web site: www.FPE6e.com For the SIMULINK files, there are equivalent LabView files that can be obtained by a link from the same web site. We have removed some material that was judged to be less useful for the teaching of a first course in controls. However, recognizing that there may still be some instructors who choose to teach the material, or students who want to refer to enrichment and/or review material, we have moved the material to the website above for access by anyone. The Table of Contents provides a guide as to where the various topics are located.

xiii

xiv Preface We feel that this Sixth Edition presents the material with good pedagogical support, provides strong motivation for the study of control, and represents a solid foundation for meeting the educational challenges. We introduce the study of feedback control, both as a specialty of itself and as support for many other fields.

Addressing the Educational Challenges Some of the educational challenges facing students of feedback control are longstanding; others have emerged in recent years. Some of the challenges remain for students across their entire engineering education; others are unique to this relatively sophisticated course. Whether they are old or new, general or particular, the educational challenges we perceived were critical to the evolution of this text. Here we will state several educational challenges and describe our approaches to each of them. •

CHALLENGE

Students must master design as well as analysis techniques.

Design is central to all of engineering and especially so to control systems. Students find that design issues, with their corresponding opportunities to tackle practical applications, particularly motivating. But students also find design problems difficult because design problem statements are usually poorly posed and lack unique solutions. Because of both its inherent importance for and its motivational effect on students, design is emphasized throughout this text so that confidence in solving design problems is developed from the start. The emphasis on design begins in Chapter 4 following the development of modeling and dynamic response. The basic idea of feedback is introduced first, showing its influence on disturbance rejection, tracking accuracy, and robustness to parameter changes. The design orientation continues with uniform treatments of the root locus, frequency response, and state variable feedback techniques. All the treatments are aimed at providing the knowledge necessary to find a good feedback control design with no more complex mathematical development than is essential to clear understanding. Throughout the text, examples are used to compare and contrast the design techniques afforded by the different design methods and, in the capstone case studies of Chapter 10, complex real-world design problems are attacked using all the methods in a unified way. •

CHALLENGE

New ideas continue to be introduced into control.

Control is an active field of research and hence there is a steady influx of new concepts, ideas, and techniques. In time, some of these elements develop to the point where they join the list of things every control engineer must know. This text is devoted to supporting students equally in their need to grasp both traditional and more modern topics. In each of our editions we have tried to give equal importance to root locus, frequency response, and state-variable methods for design. In this edition we continue to emphasize solid mastery of the underlying techniques. coupled with computer based methods for detailed calculation. We also provide an early introduction to data

Preface

xv

sampling and discrete controllers in recognition of the major role played by digital controllers in our field. While this material can be skipped to save time without harm to the flow of the text, we feel that it is very important for students to understand that computer control is widely used and that the most basic techniques of computer control are easily mastered. •

CHALLENGE

Students need to manage a great deal of information.

The vast array of systems to which feedback control is applied and the growing variety of techniques available for the solution of control problems means that today’s student of feedback control must learn many new ideas. How do students keep their perspective as they plow through lengthy and complex textual passages? How do they identify highlights and draw appropriate conclusions? How do they review for exams? Helping students with these tasks was a criterion for the Fourth and Fifth Editions and continues to be addressed in this Sixth Edition. We outline these features below. FEATURE 1. Chapter openers offer perspective and overview. They place the specific chapter topic in the context of the discipline as a whole and they briefly overview the chapter sections. 2. Margin notes help students scan for chapter highlights. They point to important definitions, equations, and concepts. 3. Boxed highlights identify key concepts within the running text. They also function to summarize important design procedures. 4. Bulleted chapter summaries help with student review and prioritization. These summaries briefly reiterate the key concepts and conclusions of the chapter. 5. Synopsis of design aids. Relationships used in design and throughout the book are collected inside the back cover for easy reference. 6. The color blue is used (1) to highlight useful pedagogical features, (2) to highlight components under particular scrutiny within block diagrams, (3) to distinguish curves on graphs, and (4) to lend a more realistic look to figures of physical systems. 7. Review questions at the end of each chapter with solutions in the back to guide the student in self-study •

CHALLENGE disciplines.

Students of feedback control come from a wide range of

Feedback control is an interdisciplinary field in that control is applied to systems in every conceivable area of engineering. Consequently, some schools have separate introductory courses for control within the standard disciplines and some, like Stanford, have a single set of courses taken by students from many disciplines. However, to restrict the examples to one field is to miss much of the range and power of feedback but to cover the whole range of applications is overwhelming. In this book we develop the interdisciplinary nature of the field and provide review material for several of the

xvi Preface most common technologies so that students from many disciplines will be comfortable with the presentation. For Electrical Engineering students who typically have a good background in transform analysis, we include in Chapter 2 an introduction to writing equations of motion for mechanical mechanisms. For mechanical engineers, we include in Chapter 3 a review of the Laplace Transform and dynamic response as needed in control. In addition, we introduce other technologies briefly and, from time to time, we present the equations of motion of a physical system without derivation but with enough physical description to be understood from a response point of view. Examples of some of the physical systems represented in the text include the read– write head for a computer disk drive, a satellite tracking system, the fuel–air ratio in an automobile engine, and an airplane automatic pilot system.

Outline of the Book The contents of the book are organized into ten chapters and three appendixes. Optional sections of advanced or enrichment material marked with a triangle () are included at the end of some chapters. There is additional enrichment material on the website. Examples and problems based on this material are also marked with a triangle (). The appendices include background and reference material. The appendices in the book include Laplace transform tables, answers to the end-of-chapter review questions, and a list of MATLAB commands. The appendixes on the website include a review of complex variables, a review of matrix theory, some important results related to State-Space design, a tutorial on RLTOOL for MATLAB, and optional material supporting or extending several of the chapters. In Chapter 1, the essential ideas of feedback and some of the key design issues are introduced. This chapter also contains a brief history of control, from the ancient beginnings of process control to flight control and electronic feedback amplifiers. It is hoped that this brief history will give a context for the field, introduce some of the key figures who contributed to its development, and provide motivation to the student for the studies to come. Chapter 2 is a short presentation of dynamic modeling and includes mechanical, electrical, electromechanical, fluid, and thermodynamic devices. This material can be omitted, used as the basis of review homework to smooth out the usual nonuniform preparation of students, or covered in-depth depending on the needs of the students. Chapter 3 covers dynamic response as used in control. Again, much of this material may have been covered previously, especially by electrical engineering students. For many students, the correlation between pole locations and transient response and the effects of extra zeros and poles on dynamic response represent new material. Stability of dynamic systems is also introduced in this Chapter. This material needs to be covered carefully. Chapter 4 presents the basic equations and transfer functions of feedback along with the definitions of the sensitivity function. With these tools, open-loop and closedloop control are compared with respect to disturbance rejection, tracking accuracy, and sensitivity to model errors. Classification of systems according to their ability to track polynomial reference signals or to reject polynomial disturbances is described with the concept of system type. Finally, the classical proportional, integral, and derivative (PID) control structure is introduced and the influence of the controller

Preface

xvii

parameters on a system’s characteristic equation is explored along with PID tuning methods. The end-of-chapter optional section treats digital control. Following the overview of feedback in Chapter 4, the core of the book presents the design methods based on root locus, frequency response, and state-variable feedback in Chapters 5, 6, and 7, respectively. Chapter 8 develops in more detail the tools needed to design feedback control for implementation in a digital computer. However, for a complete treatment of feedback control using digital computers, the reader is referred to the companion text, Digital Control of Dynamic Systems, by Franklin, Powell, and Workman; Ellis-Kagle Press, 1998. In Chapter 9 the nonlinear material includes techniques for the linearization of equations of motion, analysis of zero memory nonlinearity as a variable gain, frequency response as a describing function, the phase plane, Lyapunov stability theory, and the circle stability criterion. In Chapter 10 the three primary approaches are integrated in several case studies and a framework for design is described that includes a touch of the real-world context of practical control design.

Course Configurations The material in this text can be covered flexibly. Most first-course students in controls will have some dynamics and Laplace transforms. Therefore, Chapter 2 and most of Chapter 3 would be a review for those students. In a ten-week quarter, it is possible to review Chapter 3, and all of Chapters 1, 4, 5, and 6. Most boxed sections should be omitted. In the second quarter, Chapters 7, and 9 can be covered comfortably including the boxed sections. Alternatively, some boxed sections could be omitted and selected portions of Chapter 8 included. A semester course should comfortably accommodate Chapters 1–7, including the review material of Chapters 2 and 3, if needed. If time remains after this core coverage, some introduction of digital control from Chapter 8, selected nonlinear issues from Chapter 9 and some of the case studies from Chapter 10 may be added. The entire book can also be used for a three-quarter sequence of courses consisting of modeling and dynamic response (Chapters 2 and 3), classical control (Chapters 4–6), and modern control (Chapters 7–10). Two basic 10-week courses are offered at Stanford and are taken by seniors and first-year graduate students who have not had a course in control, mostly in the departments of Aeronautics and Astronautics, Mechanical Engineering, and Electrical Engineering. The first course reviews Chapters 2 and 3 and covers Chapters 4–6. The more advanced course is intended for graduate students and reviews Chapters 4–6 and covers Chapters 7–10. This sequence complements a graduate course in linear systems and is the prerequisite to courses in digital control, nonlinear control, optimal control, flight control, and smart product design. Several of the subsequent courses include extensive laboratory experiments. Prerequisites for the course sequence include dynamics or circuit analysis and Laplace transforms.

xviii Preface

Prerequisites to This Feedback Control Course This book is for a first course at the senior level for all engineering majors. For the core topics in Chapters 4–7, prerequisite understanding of modeling and dynamic response is necessary. Many students will come into the course with sufficient background in those concepts from previous courses in physics, circuits, and dynamic response. For those needing review, Chapters 2 and 3 should fill in the gaps. An elementary understanding of matrix algebra is necessary to understand the state-space material. While all students will have much of this in prerequisite math courses, a review of the basic relations is given in Appendix WE and a brief treatment of particular material needed in control is given at the start of Chapter 7. The emphasis is on the relations between linear dynamic systems and linear algebra.

Supplements The Web site mentioned above includes the dot-m and dot-mdl files used to generate all the MATLAB figures in the book and these may be copied and distributed to the students as desired. An instructor’s manual with complete solutions to all homework problems is available. The Web site also includes advanced material and appendixes.

Acknowledgments Finally, we wish to acknowledge our great debt to all those who have contributed to the development of feedback control into the exciting field it is today and specifically to the considerable help and education we have received from our students and our colleagues. In particular, we have benefited in this effort by many discussions with the following who taught introductory control at Stanford: A. E. Bryson, Jr., R. H. Cannon, Jr., D. B. DeBra, S. Rock, S. Boyd, C. Tomlin, P. Enge, and C. Gerdes. Other colleagues who have helped us include D. Fraser, N. C. Emami, B. Silver, M. Dorfman, D. Brennan, K. Rudie, L. Pao, F. Khorrami, K. Lorell, and P. D. Mathur. Special thanks go to the many students who have provided almost all the solutions to the problems in the book. G.F.F. J.D.P. A.E.-N. Stanford, California

1

An Overview and Brief History of Feedback Control

A Perspective on Feedback Control Feedback control of dynamic systems is a very old concept with many characteristics that have evolved over time. The central idea is that a system’s output can be measured and fed back to a controller of some kind and used to effect the control. It has been shown that signal feedback can be used to control a vast array of dynamic systems including, for example, airplanes and hard-disk data storage devices. To achieve good control there are four basic requirements. • • • •

The system must be stable at all times The system output must track the command input signal The system output must be prevented from responding too much to disturbance inputs These goals must be met even if the model used in the design is not completely accurate or if the dynamics of the physical system change over time or with environmental changes.

The requirement of stability is basic and may have two causes. In the first place, the system may be unstable. This is illustrated by the Segway vehicle, which will simply fall over if the control is turned off. On the other hand, adding feedback may itself drive the system unstable. In ordinary experience

1

2 Chapter 1 An Overview and Brief History of Feedback Control such an instability is called a “vicious circle,” where the feedback signal that is circled back makes the situation worse rather than better. There are many examples of the requirement of having the system’s output track a command signal. For example, driving a car so that the vehicle stays in its lane is command tracking. Similarly, flying an airplane in the approach to a landing strip requires that a glide path be accurately tracked. Disturbance rejection is one of the very oldest applications of feedback control. In this case, the “command” is simply a constant set point to which the output is to be held as the environment changes. A very common example of this is the room thermostat whose job it is to hold the room temperature close to the set point as outside temperature and wind change, and as doors and windows are opened and closed. Finally, to design a controller for a dynamic system, it is necessary to have a mathematical model of the dynamic response of the system in all but the simplest cases. Unfortunately, almost all physical systems are very complex and often nonlinear. As a result, the design will usually be based on a simplified model and must be robust enough that the system meets its performance requirements when applied to the real device. Furthermore, again in almost all cases, as time and the environment change, even the best of models will be in error because the system dynamics have changed. Again, the design must not be too sensitive to these inevitable changes and it must work well enough regardless. The tools available to control engineers to solve these problems have evolved over time as well. Especially important has been the development of digital computers both as computation aids and as embedded control devices. As computation devices, computers have permitted identification of increasingly complex models and the application of very sophisticated control design methods. Also, as embedded devices, digital devices have permitted the implementation of very complex control laws. Control engineers must not only be skilled in manipulating these design tools but also need to understand the concepts behind these tools to be able to make the best use of them. Also important is that the control engineer understand both the capabilities and the limitations of the controller devices available.

Chapter Overview In this chapter we begin our exploration of feedback control using a simple familiar example: a household furnace controlled by a thermostat. The generic components of a control system are identified within the context of this example. In another example—an automobile cruise control—we develop the elementary static equations and assign numerical values to elements of the system model in order to compare the performance of open-loop control to that of feedback control when dynamics are ignored. In order to provide a context for our studies and to give you a glimpse of how the field has evolved, Section 1.3 provides a brief history of control theory and design. In addition, later chapters have brief sections of additional historical notes on the topics

1.1 A Simple Feedback System

3

covered there. Finally, Section 1.4 provides a brief overview of the contents and organization of the entire book.

1.1 A Simple Feedback System In feedback systems the variable being controlled—such as temperature or speed—is measured by a sensor and the measured information is fed back to the controller to influence the controlled variable. The principle is readily illustrated by a very common system, the household furnace controlled by a thermostat. The components of this system and their interconnections are shown in Fig. 1.1. Such a picture identifies the major parts of the system and shows the directions of information flow from one component to another. We can easily analyze the operation of this system qualitatively from the graph. Suppose both the temperature in the room where the thermostat is located and the Heat loss Qout Desired temperature

Gas valve

Thermostat

Furnace

Qin

Room temperature








House

(a) 70 60 Room temperature Temperature (˚F)

50 Outside temperature 40 30 20 Furnace off

Furnace on

10 0

0

2

4

6

8 10 Time (hours) (b)

12

14

16

Figure 1.1 (a) Component block diagram of a room temperature control system; (b) plot of room temperature and furnace action

4 Chapter 1 An Overview and Brief History of Feedback Control outside temperature are significantly below the reference temperature (also called the set point) when power is applied. The thermostat will be on and the control logic will open the furnace gas valve and light the fire box. This will cause heat Qin to be supplied to the house at a rate that will be significantly larger than the heat loss Qout . As a result, the room temperature will rise until it exceeds the thermostat reference setting by a small amount. At this time the furnace will be turned off and the room temperature will start to fall toward the outside value. When it falls a small amount below the set point, the thermostat will come on again and the cycle will repeat. Typical plots of room temperature along with the furnace cycles of on and off are shown in Fig. 1.1. The outside temperature is held at 50◦ F and the thermostat is initially set at 55◦ F. At 6 a.m., the thermostat is stepped to 65◦ F and the furnace brings it to that level and cycles the temperature around that figure thereafter.1 Notice that the house is well insulated, so that the fall of temperature with the furnace off is significantly slower than the rise with the furnace on. From this example, we can identify the generic components of the elementary feedback control system as shown in Fig. 1.2. The central component of this feedback system is the process whose output is to be controlled. In our example the process would be the house whose output is the room temperature and the disturbance to the process is the flow of heat from the house due to conduction through the walls and roof to the lower outside temperature. (The outward flow of heat also depends on other factors such as wind, open doors, etc.) The design of the process can obviously have a major impact on the effectiveness of the controls. The temperature of a well-insulated house with thermopane windows is clearly easier to control than otherwise. Similarly, the design of aircraft with control in mind makes a world of difference to the final performance. In every case, the earlier the issues of control are introduced into the process design, the better. The actuator is the device that can influence the controlled variable of the process and in our case, Disturbance Plant

Reference

Input filter





Control signal Controller

Output

Actuator

Process




Sensor

Sensor noise

Figure 1.2 Component block diagram of an elementary feedback control 1 Notice that the furnace had come on a few minutes before 6 a.m. on its regular nighttime schedule.

1.1 A Simple Feedback System

5

the actuator is a gas furnace. Actually, the furnace usually has a pilot light or striking mechanism, a gas valve, and a blower fan, which turns on or off depending on the air temperature in the furnace. These details illustrate the fact that many feedback systems contain components that themselves form other feedback systems.2 The central issue with the actuator is its ability to move the process output with adequate speed and range. The furnace must produce more heat than the house loses on the worst day and must distribute it quickly if the house temperature is to be kept in a narrow range. Power, speed, and reliability are usually more important than accuracy. Generally, the process and the actuator are intimately connected and the control design centers on finding a suitable input or control signal to send to the actuator. The combination of process and actuator is called the plant and the component that actually computes the desired control signal is the controller. Because of the flexibility of electrical signal processing, the controller typically works on electrical signals although the use of pneumatic controllers based on compressed air has a long and important place in process control. With the development of digital technology, cost-effectiveness and flexibility have led to the use of digital signal processors as the controller in an increasing number of cases. The component labeled thermostat in Fig. 1.1 measures the room temperature and is called the sensor in Fig. 1.2, a device whose output inevitably contains sensor noise. Sensor selection and placement are very important in control design, for it is sometimes not possible for the true controlled variable and the sensed variable to be the same. For example, although we may really wish to control the house temperature as a whole, the thermostat is in one particular room, which may or may not be at the same temperature as the rest of the house. For instance, if the thermostat is set to 68◦ F but is placed in the living room near a roaring fireplace, a person working in the study could still feel uncomfortably cold.3,4 As we will see, in addition to placement, important properties of a sensor are the accuracy of the measurements as well as low noise, reliability, and linearity. The sensor will typically convert the physical variable into an electrical signal for use by the controller. Our general system also includes an input filter whose role is to convert the reference signal to electrical form for later manipulation by the controller. In some cases the input filter can modify the reference command input in ways that improve the system response. Finally, there is a comparator to compute the difference between the reference signal and the sensor output to give the controller a measure of the system error. This text will present methods for analyzing feedback control systems and their components and will describe the most important design techniques engineers can use with confidence in applying feedback to solve control problems. We will also study 2 Jonathan Swift (1733) said it this way: “So, Naturalists observe, a flea Hath smaller fleas that on him prey; And these have smaller still to bite ‘em; And so proceed, ad infinitum.” 3 In the renovations of the kitchen in the house of one of the authors, the new ovens were placed against the wall where the thermostat was mounted on the other side. Now when dinner is baked in the kitchen on a cold day, the author freezes in his study unless the thermostat is reset. 4 The story is told of the new employee at the nitroglycerin factory who was to control the temperature of a critical part of the process manually. He was told to “keep that reading below 300◦ .” On a routine

inspection tour, the supervisor realized that the batch was dangerously hot and found the worker holding the thermometer under cold water tap to bring it down to 300◦ . They got out just before the explosion. Moral: sometimes automatic control is better than manual.

6 Chapter 1 An Overview and Brief History of Feedback Control the specific advantages of feedback that compensate for the additional complexity it demands. However, although the temperature control system is easy to understand, it is nonlinear as seen by the fact that the furnace is either on or off, and to introduce linear controls we need another example.

1.2 A First Analysis of Feedback The value of feedback can be readily demonstrated by quantitative analysis of a simplified model of a familiar system, the cruise control of an automobile (Fig. 1.3). To study this situation analytically, we need a mathematical model of our system in the form of a set of quantitative relationships among the variables. For this example, we ignore the dynamic response of the car and consider only the steady behavior. (Dynamics will, of course, play a major role in later chapters.) Furthermore, we assume that for the range of speeds to be used by the system, we can approximate the relations as linear. After measuring the speed of the vehicle on a level road at 65 mph, we find that a 1◦ change in the throttle angle (our control variable) causes a 10-mph change in speed. From observations while driving up and down hills it is found that when the grade changes by 1%, we measure a speed change of 5 mph. The speedometer is found to be accurate to a fraction of 1 mph and will be considered exact. With these relations, we can draw the block diagram of the plant (Fig. 1.4), which shows these mathematical relationships in graphical form. In this diagram the connecting lines carry signals and a block is like an ideal amplifier which multiplies the signal at its input by the value marked in the block to give the output signal. To sum two or more signals, we show lines for the signals coming into a summer, a circle with the summation sign
inside. An algebraic sign (plus or minus) beside each arrow head indicates whether the input adds to or subtracts from the total output of the summer. For this analysis, we wish to compare the effects of a 1% grade on the output speed when the reference speed is set for 65 with and without feedback to the controller. Figure 1.3 Component block diagram of automobile cruise control

Road grade Controller Desired speed

Control variable

Actuator

Process Auto body

Engine

?? Throttle

Sensor

Measured speed

Speedometer

Sensor noise

Actual speed

1.2 A First Analysis of Feedback Figure 1.4 Block diagram of the cruise control plant

7

w (% grade)

0.5

Control (°) u

Output speed (mph)






10

Figure 1.5 Open-loop cruise control

w

y

Plant

0.5 Controller r

Open-loop control

1/10


u 



10

y

In the first case, shown in Fig. 1.5, the controller does not use the speedometer reading but sets u = r/10. This is an example of an open-loop control system. The term open loop refers to the fact that there is no closed path or loop around which the signals go in the block diagram. In our simple example the open-loop output speed, yol , is given by the equations yol = 10(u − 0.5w) 
r − 0.5w = 10 10 = r − 5w. The error in output speed is

and the percent error is

eol = r − yol

(1.1)

= 5w

(1.2)

w (1.3) % error = 500 . r If r = 65 and the road is level then w = 0 and the speed will be 65 with no error. However, if w = 1 corresponding to a 1% grade, then the speed will be 60 and we have a 5-mph error, which is a 7.69% error in the speed. For a grade of 2%, the speed error would be 10 mph, which is an error of 15.38%, and so on. The example shows that there would be no error when w = 0 but this result depends on the controller gain being the exact inverse of the plant gain of 10. In practice, the plant gain is subject to change and if it does, errors are introduced by this means also. If there is an error in the plant gain in open-loop control, the percent speed error would be the same as the percent plant-gain error.

8 Chapter 1 An Overview and Brief History of Feedback Control The block diagram of a feedback scheme is shown in Fig. 1.6, where the controller gain has been set to 10. Recall that in this simple example, we have assumed that we have an ideal sensor whose block is not shown. In this case the equations are ycl = 10u − 5w, u = 10(r − ycl ). Combining them yields ycl = 100r − 100ycl − 5w, 101ycl = 100r − 5w, 5 100 r− w, 101 101 5w r + . ecl = 101 101 Thus the feedback has reduced the sensitivity of the speed error to the grade by a factor of 101 when compared with the open-loop system. Note, however, that there is now a small speed error on level ground because even when w = 0, 100 ycl = r = 0.99r mph. 101 This error will be small as long as the loop gain (product of plant and controller gains) is large.5 If we again consider a reference speed of 65 mph and compare speeds with a 1% grade, the percent error in the output speed is   65 × 100 65 × 100 5 − − 101 101 101 % error = 100 (1.4) 65 × 100 101 5 × 101 (1.5) = 100 101 × 65 × 100 = 0.0769%. (1.6) ycl =

Figure 1.6 Closed-loop cruise control

w

Plant

0.5 Controller

r






10




u 



10

ycl

5 In case the error is too large, it is common practice to reset the reference, in this case to 101 r, so the 100 output reaches the true desired value.

1.3 A Brief History

The design trade-off

9

The reduction of the speed sensitivity to grade disturbances and plant gain in our example is due to the loop gain of 100 in the feedback case. Unfortunately, there are limits to how high this gain can be made; when dynamics are introduced, the feedback can make the response worse than before, or even cause the system to become unstable. The dilemma is illustrated by another familiar situation where it is easy to change a feedback gain. If one tries to raise the gain of a public-address amplifier too much, the sound system will squeal in a most unpleasant way. This is a situation where the gain in the feedback loop—from the speakers to the microphone through the amplifier back to the speakers—is too much. The issue of how to get the gain as large as possible to reduce the errors without making the system become unstable and squeal is what much of feedback control design is all about.

1.3 A Brief History

Liquid-level control

Drebbel’s incubator

Figure 1.7 Early historical control of liquid level and flow

An interesting history of early work on feedback control has been written by O. Mayr (1970), who traces the control of mechanisms to antiquity. Two of the earliest examples are the control of flow rate to regulate a water clock and the control of liquid level in a wine vessel, which is thereby kept full regardless of how many cups are dipped from it. The control of fluid flow rate is reduced to the control of fluid level, since a small orifice will produce constant flow if the pressure is constant, which is the case if the level of the liquid above the orifice is constant. The mechanism of the liquid-level control invented in antiquity and still used today (for example, in the water tank of the ordinary flush toilet) is the float valve. As the liquid level falls, so does the float, allowing the flow into the tank to increase; as the level rises, the flow is reduced and if necessary cut off. Figure 1.7 shows how a float valve operates. Notice here that sensor and actuator are not separate devices but are contained in the carefully shaped float-and-supply-tube combination. A more recent invention described by Mayr (1970) is a system, designed by Cornelis Drebbel in about 1620, to control the temperature of a furnace used to heat an incubator6 (Fig. 1.8). The furnace consists of a box to contain the fire, with a flue at the top fitted with a damper. Inside the fire box is the double-walled incubator box, the hollow walls of which are filled with water to transfer the heat evenly to the incubator. The temperature sensor is a glass vessel filled with alcohol and mercury and placed in the water jacket around the incubator box. As the fire heats the box and Supply

Float

6 French doctors introduced incubators into the care of premature babies over 100 years ago.

10 Chapter 1 An Overview and Brief History of Feedback Control Figure 1.8 Drebbel’s incubator for hatching chicken eggs

Damper

Source: Adapted from Mayr, 1970

Riser

Eggs Float

Water

Flue gases Mercury Metal plate

Fly-ball governor

Fire

Alcohol

water, the alcohol expands and the riser floats up, lowering the damper on the flue. If the box is too cold, the alcohol contracts, the damper is opened, and the fire burns hotter. The desired temperature is set by the length of the riser, which sets the opening of the damper for a given expansion of the alcohol. A famous problem in the chronicles of control systems was the search for a means to control the rotation speed of a shaft. Much early work (Fuller, 1976) seems to have been motivated by the desire to automatically control the speed of the grinding stone in a wind-driven flour mill. Of various methods attempted, the one with the most promise used a conical pendulum, or fly-ball governor, to measure the speed of the mill. The sails of the driving windmill were rolled up or let out with ropes and pulleys, much like a window shade, to maintain fixed speed. However, it was adaptation of these principles to the steam engine in the laboratories of James Watt around 1788 that made the fly-ball governor famous. An early version is shown in Fig. 1.9, while Figs. 1.10 and 1.11 show a close-up of a fly-ball governor and a sketch of its components. The action of the fly-ball governor (also called a centrifugal governor) is simple to describe. Suppose the engine is operating in equilibrium. Two weighted balls spinning around a central shaft can be seen to describe a cone of a given angle with the shaft. When a load is suddenly applied to the engine, its speed will slow, and the balls of the governor will drop to a smaller cone. Thus the ball angle is used to sense the output speed. This action, through the levers, will open the main valve to the steam chest (which is the actuator) and admit more steam to the engine, restoring most of the lost speed. To hold the steam valve at a new position it is necessary for the fly balls to rotate at a different angle, implying that the speed under load is not exactly the same as before. We saw this effect earlier with cruise control, where feedback control gave a very small error. To recover the exact same speed in the system, it would require resetting the desired speed setting by changing the length of the rod from the lever to the valve. Subsequent inventors introduced mechanisms that integrated the speed error to provide automatic reset. In Chapter 4 we will analyze these systems to show

1.3 A Brief History

11

Figure 1.9 Photograph of an early Watt steam engine Source: British Crown Copyright, Science Museum, London

Figure 1.10 Close-up of the fly-ball governor Source: British Crown Copyright, Science Museum, London

Beginnings of control theory

that such integration can result in feedback systems with zero steady-state error to constant disturbances. Because Watt was a practical man, like the millwrights before him, he did not engage in theoretical analysis of the governor. Fuller (1976) has traced the early development of control theory to a period of studies from Christian Huygens in 1673 to James Clerk Maxwell in 1868. Fuller gives particular credit to the contributions of G. B. Airy, professor of mathematics and astronomy at Cambridge University from 1826 to 1835 and Astronomer Royal at Greenwich Observatory from 1835 to 1881. Airy was concerned with speed control; if his telescopes could be rotated counter to the rotation of the earth, a fixed star could be observed for extended periods. Using the centrifugal-pendulum governor he discovered that it was capable of unstable motion— “and the machine (if I may so express myself) became perfectly wild” (Airy, 1840; quoted in Fuller, 1976). According to Fuller, Airy was the first worker to discuss

12 Chapter 1 An Overview and Brief History of Feedback Control Figure 1.11 Operating parts of a fly-ball governor

Pivot

Sleeve Butterfly valve To engine inlet

Steam

Balls

Rotation Pulley from engine

Stability analysis

instability in a feedback control system and the first to analyze such a system using differential equations. These attributes signal the beginnings of the study of feedback control dynamics. The first systematic study of the stability of feedback control was apparently given in the paper “On Governors” by J. C. Maxwell (1868).7 In this paper, Maxwell developed the differential equations of the governor, linearized them about equilibrium, and stated that stability depends on the roots of a certain (characteristic) equation having negative real parts. Maxwell attempted to derive conditions on the coefficients of a polynomial that would hold if all the roots had negative real parts. He was successful only for second- and third-order cases. Determining criteria for stability was the problem for the Adams Prize of 1877, which was won by E. J. Routh.8 His criterion, developed in his essay, remains of sufficient interest that control engineers are still learning how to apply his simple technique. Analysis of the characteristic equation remained the foundation of control theory until the invention of the electronic feedback amplifier by H. S. Black in 1927 at Bell Telephone Laboratories. Shortly after publication of Routh’s work, the Russian mathematician A. M. Lyapunov (1893) began studying the question of stability of motion. His studies were based on the nonlinear differential equations of motion and also included results for linear equations that are equivalent to Routh’s criterion. His work was fundamental to what is now called the state-variable approach to control theory, but was not introduced into the control literature until about 1958. 7An exposition of Maxwell’s contribution is given in Fuller (1976). 8 E. J. Routh was first academically in his class at Cambridge University in 1854, while J. C. Maxwell was

second. In 1877 Maxwell was on the Adams Prize Committee that chose the problem of stability as the topic for the year.

1.3 A Brief History

Frequency response

PID control

13

The development of the feedback amplifier is briefly described in an interesting article based on a talk by H. W. Bode (1960) reproduced in Bellman and Kalaba (1964). With the introduction of electronic amplifiers, long-distance telephoning became possible in the decades following World War I. However, as distances increased, so did the loss of electrical energy; in spite of using larger-diameter wires, increasing numbers of amplifiers were needed to replace the lost energy. Unfortunately, large numbers of amplifiers resulted in much distortion since the small nonlinearity of the vacuum tubes then used in electronic amplifiers were multiplied many times. To solve the problem of reducing distortion, Black proposed the feedback amplifier. As mentioned earlier in connection with the automobile cruise control, the more we wish to reduce errors (or distortion), the more feedback we need to apply. The loop gain from actuator to plant to sensor to actuator must be made very large. With high gain the feedback loop begins to squeal and is unstable. Here was Maxwell’s and Routh’s stability problem again, except that in this technology the dynamics were so complex (with differential equations of order 50 being common) that Routh’s criterion was not very helpful. So the communications engineers at Bell Telephone Laboratories, familiar with the concept of frequency response and the mathematics of complex variables, turned to complex analysis. In 1932 H. Nyquist published a paper describing how to determine stability from a graphical plot of the loop frequency response. From this theory there developed an extensive methodology of feedback-amplifier design described by Bode (1945) and extensively used still in the design of feedback controls. Nyquist and Bode plots are discussed in more detail in Chapter 6. Simultaneous with the development of the feedback amplifier, feedback control of industrial processes was becoming standard. This field, characterized by processes that are not only highly complex but also nonlinear and subject to relatively long time delays between actuator and sensor, developed proportional-integral-derivative (PID) control. The PID controller was first described by Callender et al. (1936). This technology was based on extensive experimental work and simple linearized approximations to the system dynamics. It led to standard experiments suitable to application in the field and eventually to satisfactory “tuning” of the coefficients of the PID controller. (PID controllers are covered in Chapter 4.) Also under development at this time were devices for guiding and controlling aircraft; especially important was the development of sensors for measuring aircraft altitude and speed. An interesting account of this branch of control theory is given in McRuer (1973). An enormous impulse was given to the field of feedback control during World War II. In the United States engineers and mathematicians at the MIT Radiation Laboratory combined their knowledge to bring together not only Bode’s feedback amplifier theory and the PID control of processes but also the theory of stochastic processes developed by N. Wiener (1930). The result was the development of a comprehensive set of techniques for the design of servomechanisms, as control mechanisms came to be called. Much of this work was collected and published in the records of the Radiation Laboratory by James et al. (1947). Another approach to control systems design was introduced in 1948 by W. R. Evans, who was working in the field of guidance and control of aircraft. Many of his problems involved unstable or neutrally stable dynamics, which made the frequency methods difficult, so he suggested returning to the study of the characteristic equation that had been the basis of the work of Maxwell and Routh nearly 70 years

14 Chapter 1 An Overview and Brief History of Feedback Control

Root locus

State-variable design

Modern control Classical control

earlier. However, Evans developed techniques and rules allowing one to follow graphically the paths of the roots of the characteristic equation as a parameter was changed. His method, the root locus, is suitable for design as well as for stability analysis and remains an important technique today. The root-locus method developed by Evans is covered in Chapter 5. During the 1950s several authors, including R. Bellman and R. E. Kalman in the United States and L. S. Pontryagin in the U.S.S.R., began again to consider the ordinary differential equation (ODE) as a model for control systems. Much of this work was stimulated by the new field of control of artificial earth satellites, in which the ODE is a natural form for writing the model. Supporting this endeavor were digital computers, which could be used to carry out calculations unthinkable 10 years before. (Now, of course, these calculations can be done by any engineering student with a desktop computer.) The work of Lyapunov was translated into the language of control at about this time, and the study of optimal controls, begun by Wiener and Phillips during World War II, was extended to optimizing trajectories of nonlinear systems based on the calculus of variations. Much of this work was presented at the first conference of the newly formed International Federation of Automatic Control held in Moscow in 1960.9 This work did not use the frequency response or the characteristic equation but worked directly with the ODE in “normal” or “state” form and typically called for extensive use of computers. Even though the foundations of the study of ODEs were laid in the late 19th century, this approach is now often called modern control to distinguish it from classical control, which uses the complex variable methods of Bode and others. In the period from the 1970s continuing through the present, we find a growing body of work that seeks to use the best features of each technique. Thus we come to the current state of affairs where the principles of control are applied in a wide range of disciplines, including every branch of engineering. The well-prepared control engineer needs to understand the basic mathematical theory that underlies the field and must be able to select the best design technique suited to the problem at hand. With the ubiquitous use of computers it is especially important that the engineer is able to use his or her knowledge to guide and verify calculations done on the computer.10

1.4 An Overview of the Book The central purpose of this book is to introduce the most important techniques for single-input–single-output control systems design. Chapter 2 will review the techniques necessary to obtain models of the dynamic systems that we wish to control. These include model making for mechanical, electric, electromechanical, and a few other physical systems. Chapter 2 also describes briefly the linearization of nonlinear models, although this will be discussed more thoroughly in Chapter 9. 9 Optimal control gained a large boost when Bryson and Denham (1962) showed that the path of a supersonic

aircraft should actually dive at one point in order to reach a given altitude in minimum time. This nonintuitive result was later demonstrated to skeptical fighter pilots in flight tests. 10 For more background on the history of control, see the survey papers appearing in the IEEE Control Systems Magazine of November 1984 and June 1996.

1.4 An Overview of the Book

Computer aids

15

In Chapter 3 and Appendix A we will discuss the analysis of dynamic response using Laplace transforms along with the relationship between time response and the poles and zeros of a transfer function. The chapter also includes a discussion of the critical issue of system stability, including the Routh test. In Chapter 4 we will cover the basic equations and features of feedback. An analysis of the effects of feedback on disturbance rejection, tracking accuracy, sensitivity to parameter changes, and dynamic response will be given. The idea of elementary PID control is discussed. Also in this chapter a brief introduction is given to the digital implementation of transfer functions and thus of linear time-invariant controllers so that the effects of digital control can be compared with analog controllers as these are designed. In Chapters 5, 6, and 7 we introduce the techniques for realizing the control objectives first identified in Chapter 4 in more complex dynamic systems. These methods include the root locus, frequency response, and state-variable techniques. These are alternative means to the same end and have different advantages and disadvantages as guides to design of controls. The methods are fundamentally complementary, and each needs to be understood to achieve the most effective control systems design. In Chapter 8 we develop further the ideas of implementing controllers in a digital computer that were introduced in Chapter 4. The chapter addresses how one “digitizes” the control equations developed in Chapters 5 through 7, how the sampling introduces a delay that tends to destabilize the system, and how the sample rate needs to be a certain multiple of the system frequencies for good performance. The analysis of sampled systems requires another analysis tool—the z-transform—and that tool is described and its use is illustrated. Most real systems are nonlinear to some extent. However, the analyses and design methods in most of the book up to here are for linear systems. In Chapter 9 we explain why the study of linear systems is pertinent, why it is useful for design even though most systems are nonlinear, and how designs for linear systems can be modified to handle most common nonlinearities in the systems being controlled. The chapter covers saturation, describing functions and the anti windup controller, and contains a brief introduction to Lyapunov stability theory. Application of all the techniques to problems of substantial complexity are discussed in Chapter 10, in which the design methods are brought to bear simultaneously on specific case studies. Control designers today make extensive use of computer-aided control systems design software that is commercially available. Furthermore, most instructional programs in control systems design make software tools available to the students. The most widely used software for the purpose are MATLAB
and SIMULINK
from The Mathworks. MATLAB routines have been included throughout the text to help illustrate this method of solution and many problems require computer aids for solution. Many of the figures in the book were created using MATLAB and the files for their creation are available free of charge on the web at the site: http://www.FPE6e.com. Students and instructors are invited to use these files as it is believed that they should be helpful in learning how to use computer methods to solve control problems. Needless to say, many topics are not treated in the book. We do not extend the methods to multivariable controls, which are systems with more than one input and/or

16 Chapter 1 An Overview and Brief History of Feedback Control output, except as part of the case study of the rapid thermal processor in Chapter 10. Nor is optimal control treated in more than a very introductory manner in Chapter 7. Also beyond the scope of this text is a detailed treatment of the experimental testing and modeling of real hardware, which is the ultimate test of whether any design really works. The book concentrates on analysis and design of linear controllers for linear plant models—not because we think that is the final test of a design, but because that is the best way to grasp the basic ideas of feedback and is usually the first step in arriving at a satisfactory design. We believe that mastery of the material here will provide a foundation of understanding on which to build knowledge of these more advanced and realistic topics—a foundation strong enough to allow one to build a personal design method in the tradition of all those who worked to give us the knowledge we present here.

SUMMARY •

•

•

•

•

Control is the process of making a system variable adhere to a particular value, called the reference value. A system designed to follow a changing reference is called tracking control or a servo. A system designed to maintain an output fixed regardless of the disturbances present is called a regulating control or a regulator. Two kinds of control were defined and illustrated based on the information used in control and named by the resulting structure. In open-loop control the system does not measure the output and there is no correction of the actuating signal to make that output conform to the reference signal. In closed-loop control the system includes a sensor to measure the output and uses feedback of the sensed value to influence the control variable. A simple feedback system consists of the process whose output is to be controlled, the actuator whose output causes the process output to change, reference and output sensors that measure these signals, and the controller that implements the logic by which the control signal that commands the actuator is calculated. Block diagrams are helpful for visualizing system structure and the flow of information in control systems. The most common block diagrams represent the mathematical relationships among the signals in a control system. The theory and design techniques of control have come to be divided into two categories: classical control methods use the Laplace or Fourier Transforms and were the dominant methods for control design until about 1960 while modern control methods are based on ODEs in state form and were introduced into the field starting in the1960s. Many connections have been discovered between the two categories and well prepared engineers must be familiar with both techniques.

REVIEW QUESTIONS 1. What are the main components of a feedback control system? 2. What is the purpose of the sensor? 3.

Give three important properties of a good sensor.

Problems

17

4. What is the purpose of the actuator? 5.

Give three important properties of a good actuator.

6. What is the purpose of the controller? Give the input(s) and output(s) of the controller. 7. What physical variable(s) of a process can be directly measured by a Hall effect sensor? 8. What physical variable is measured by a tachometer? 9.

Describe three different techniques for measuring temperature.

10. Why do most sensors have an electrical output, regardless of the physical nature of the variable being measured?

PROBLEMS 1.1

Draw a component block diagram for each of the following feedback control systems. (a) (b) (c) (d)

The manual steering system of an automobile Drebbel’s incubator The water level controlled by a float and valve Watt’s steam engine with fly-ball governor In each case, indicate the location of the elements listed below and give the units associated with each signal. • • • • • • • • •

the process the process desired output signal the sensor the actuator the actuator output signal the controller the controller output signal the reference signal the error signal

Notice that in a number of cases the same physical device may perform more than one of these functions. 1.2

Identify the physical principles and describe the operation of the thermostat in your home or office.

1.3 A machine for making paper is diagrammed in Fig. 1.12. There are two main parameters under feedback control: the density of fibers as controlled by the consistency of the thick stock that flows from the headbox onto the wire, and the moisture content of the final product that comes out of the dryers. Stock from the machine chest is diluted by white water returning from under the wire as controlled by a control valve (CV).A meter supplies a reading of the consistency. At the “dry end” of the machine, there is a moisture sensor. Draw a signal graph and identify the nine components listed in Problem 1.1 part (d) for (a) control of consistency (b) control of moisture 1.4

Many variables in the human body are under feedback control. For each of the following controlled variables, draw a graph showing the process being controlled, the sensor that measures the variable, the actuator that causes it to increase and/or decrease, the

18 Chapter 1 An Overview and Brief History of Feedback Control Storage chest

Refiner

Screens Headbox and cleaners

Machine chest

Presses

Drying section

Reel

Wire Thick stock

CV White water

Consistency meter

Moisture meter

Controller

Figure 1.12 A papermaking machine Source: From Åström (1970, p. 192); reprinted with permission

information path that completes the feedback path, and the disturbances that upset the variable. You may need to consult an encyclopedia or textbook on human physiology for information on this problem. (a) (b) (c) (d) (e)

blood pressure blood sugar concentration heart rate eye-pointing angle eye-pupil diameter

1.5 Draw a graph of the components for temperature control in a refrigerator or automobile air-conditioning system. 1.6 Draw a graph of the components for an elevator-position control. Indicate how you would measure the position of the elevator car. Consider a combined coarse and fine measurement system. What accuracies do you suggest for each sensor? Your system should be able to correct for the fact that in elevators for tall buildings there is significant cable stretch as a function of cab load. 1.7 Feedback control requires being able to sense the variable being controlled. Because electrical signals can be transmitted, amplified, and processed easily, often we want to have a sensor whose output is a voltage or current proportional to the variable being measured. Describe a sensor that would give an electrical output proportional to: (a) (b) (c) (d) (e) (f) (g)

temperature pressure liquid level flow of liquid along a pipe (or blood along an artery) force linear position rotational position linear velocity

Problems

19

(h) rotational speed (i) translational acceleration (j) torque 1.8

Each of the variables listed in Problem 1.7 can be brought under feedback control. Describe an actuator that could accept an electrical input and be used to control the variables listed. Give the units of the actuator output signal.

2

Dynamic Models

A Perspective on Dynamic Models The overall goal of feedback control is to use the principle of feedback to cause the output variable of a dynamic process to follow a desired reference variable accurately, regardless of the reference variable’s path and regardless of any external disturbances or any changes in the dynamics of the process. This complex goal is met as the result of a number of simple, distinct steps. The first of these is to develop a mathematical description (called a dynamic model) of the process to be controlled. The term model, as it is used and understood by control engineers, means a set of differential equations that describe the dynamic behavior of the process. A model can be obtained using principles of the underlying physics or by testing a prototype of the device, measuring its response to inputs, and using the data to construct an analytical model. We will focus only on using physics in this chapter. There are entire books written on experimentally determining models, sometimes called System Identification, and these techniques are described very briefly in Chapter 3. A careful control system designer will typically rely on at least some experiments to verify the accuracy of the model when it is derived from physical principles. In many cases the modeling of complex processes is difficult and expensive, especially when the important steps of building and testing prototypes are included. However, in this introductory text, we will focus on the most basic principles of modeling for the most common physical systems. More comprehensive sources and specialized texts will be referenced throughout the text where appropriate for those wishing more detail. In later chapters we will explore a variety of analysis methods for dealing with the equations of motion and their solution for purposes of designing feedback control systems.

20

2.1 Dynamics of Mechanical Systems

21

Chapter Overview The fundamental step in building a dynamic model is writing the equations of motion for the system. Through discussion and a variety of examples, Section 2.1 demonstrates how to write the equations of motion for a variety of mechanical systems. In addition, the section demonstrates the use of MATLAB® to find the time response of a simple system to a step input. Furthermore, the ideas of transfer functions and block diagrams are introduced, along with the idea that problems can also be solved via SIMULINK® . Electric circuits and electromechanical systems are modeled in Sections 2.2 and 2.3, respectively. For those wanting modeling examples for more diverse dynamic systems, Section 2.4, which is optional, extends the discussion to heat and fluid-flow systems. The chapter concludes with Section 2.5, a discussion of the history behind the discoveries that led to the knowledge that we take for granted today. The differential equations developed in modeling are often nonlinear. Because nonlinear systems are significantly more challenging to solve than linear ones and because linear models are usually adequate, the emphasis in the early chapters is primarily on linear systems. However, we do show how to linearize simple nonlinearities here in Chapter 2 and show how to use SIMULINK to numerically solve for the motion of a nonlinear system. A much more extensive discussion of linearization and analysis of nonlinear systems is contained in Chapter 9. In order to focus on the important first step of developing mathematical models, we will defer explanation of the computational methods used to solve the equations of motion developed in this chapter until Chapter 3.

2.1 Dynamics of Mechanical Systems 2.1.1 Translational Motion Newton’s law for translational motion

The cornerstone for obtaining a mathematical model, or the equations of motion, for any mechanical system is Newton’s law, F = ma,

(2.1)

where F = the vector sum of all forces applied to each body in a system, newtons (N) or pounds (lb), a = the vector acceleration of each body with respect to an inertial reference frame (i.e., one that is neither accelerating nor rotating with respect to the stars); often called inertial acceleration, m/sec2 or ft/sec2 , m = mass of the body, kg or slug. Note that here in Eq. (2.1), as throughout the text, we use the convention of boldfacing the type to indicate that the quantity is a matrix or vector, possibly a vector function.

22 Chapter 2 Dynamic Models Figure 2.1 Cruise control model

u

Use of free-body diagram in applying Newton’s law

EXAMPLE 2.1

In SI units a force of 1 N will impart an acceleration of 1 m/sec2 to a mass of 1 kg. In English units a force of 1 lb will impart an acceleration of 1 ft/sec2 to a mass of 1 slug. The “weight” of an object is mg, where g is the acceleration of gravity (= 9.81 m/sec2 = 32.2 ft/sec2 ). In English units it is common usage to refer to the mass of an object in terms of its weight in pounds, which is the quantity measured on scales. To obtain the mass in slugs for use in Newton’s law, divide the weight by g. Therefore, an object weighing 1 lb has a mass of 1/32.2 slugs. A slug has units lb·sec2 /ft. In metric units, scales are typically calibrated in kilograms, which is a direct measure of mass. Application of this law typically involves defining convenient coordinates to account for the body’s motion (position, velocity, and acceleration), determining the forces on the body using a free-body diagram, and then writing the equations of motion from Eq. (2.1). The procedure is simplest when the coordinates chosen express the position with respect to an inertial frame because, in this case, the accelerations needed for Newton’s law are simply the second derivatives of the position coordinates.

A Simple System; Cruise Control Model 1. Write the equations of motion for the speed and forward motion of the car shown in Fig. 2.1 assuming that the engine imparts a force u as shown. Take the Laplace transform of the resulting differential equation and find the transfer function between the input u and the output v. 2. Use MATLAB to find the response of the velocity of the car for the case in which the input jumps from being u = 0 at time t = 0 to a constant u = 500 N thereafter. Assume that the car mass m is 1000 kg and viscous drag coefficient, b = 50 N·sec/m. Solution 1. Equations of motion: For simplicity we assume that the rotational inertia of the wheels is negligible and that there is friction retarding the motion of the car that is proportional to the car’s speed with a proportionality constant, b.1 The car can 1 If the speed is v, the aerodynamic friction force is proportional to v2 . In this simple model we have taken a linear approximation.

2.1 Dynamics of Mechanical Systems Figure 2.2 Free-body diagram for cruise control

23

x x Friction force bx

m

u

then be approximated for modeling purposes using the free-body diagram seen in Fig. 2.2, which defines coordinates, shows all forces acting on the body (heavy lines), and indicates the acceleration (dashed line). The coordinate of the car’s position x is the distance from the reference line shown and is chosen so that positive is to the right. Note that in this case the inertial acceleration is simply the second derivative of x (i.e., a = x¨ ) because the car position is measured with respect to an inertial reference. The equation of motion is found using Eq. (2.1). The friction force acts opposite to the direction of motion; therefore it is drawn opposite the direction of positive motion and entered as a negative force in Eq. (2.1). The result is u − b˙x = m¨x ,

(2.2)

or

u b x˙ = . (2.3) m m For the case of the automotive cruise control where the variable of interest is the speed, v (=˙x ), the equation of motion becomes b u v˙ + v = . (2.4) m m The solution of such an equation will be covered in detail in Chapter 3; however, the essence is that you assume a solution of the form v = Vo est given an input of the form u = Uo est . Then, since v˙ = sVo est , the differential equation can be written as   1 b (2.5) s+ Vo est = Uo est . m m The est term cancels out, and we find that x¨ +

1 Vo = mb. Uo s+ m

(2.6)

For reasons that will become clear in Chapter 3, this is usually written as 1 V (s) = mb. U(s) s+ m

Transfer function

(2.7)

This expression of the differential equation (2.4) is called the transfer function and will be used extensively in later chapters. Note that, in essence, we have substituted s for d/dt in Eq. (2.4).2 2 The use of an operator for differentiation was developed by Cauchy about 1820 based on the Laplace

transform, which was developed about 1780. In Chapter 3 we will show how to derive transfer functions using the Laplace transform. Reference: Gardner and Barnes, 1942.

24 Chapter 2 Dynamic Models 2. Time response: The dynamics of a system can be prescribed to MATLAB in terms of row vectors containing the coefficients of the polynomials describing the numerator and denominator of its transfer function. The transfer function for this problem is that given in part (a). In this case, the numerator (called num) is simply one number since there are no powers of s, so that num = 1/m = 1/1000. The denominator (called den) contains the coefficients of the polynomial s+ b/m, which are     50 b den = 1 = 1 . m 1000 The step function in MATLAB calculates the time response of a linear system to a unit step input. Because the system is linear, the output for this case can be multiplied by the magnitude of the input step to derive a step response of any amplitude. Equivalently, num can be multiplied by the magnitude of the input step. The statements

Step response with MATLAB

num = 1/1000; den = [1 50/1000]; sys = tf(num*500, den); step(sys);

% 1/m % s + b/m % step gives unit step response, so num*500 gives u = 500. % plots the step response

calculate and plot the time response for an input step with a 500-N magnitude. The step response is shown in Fig. 2.3. Newton’s law also can be applied to systems with more than one mass. In this case it is particularly important to draw the free-body diagram of each mass, showing the applied external forces as well as the equal and opposite internal forces that act from each mass on the other.

Figure 2.3 Response of the car velocity to a step in u

10

Amplitude

8 6 4 2 0

0

50 Time (sec)

100

2.1 Dynamics of Mechanical Systems

EXAMPLE 2.2

25

A Two-Mass System: Suspension Model Figure 2.4 shows an automobile suspension system. Write the equations of motion for the automobile and wheel motion assuming one-dimensional vertical motion of one quarter of the car mass above one wheel. A system consisting of one of the four wheel suspensions is usually referred to as a quarter-car model. Assume that the model is for a car with a mass of 1580 kg, including the four wheels, which have a mass of 20 kg each. By placing a known weight (an author) directly over a wheel and measuring the car’s deflection, we find that ks = 130,000 N/m. Measuring the wheel’s deflection for the same applied weight, we find that kw  1,000,000 N/m. By using the results in Section 3.3, Fig. 3.18(b), and qualitatively observing that the car’s response as the author jumps off matches the ζ = 0.7 curve, we conclude that b = 9800 N·sec/m. Solution. The system can be approximated by the simplified system shown in Fig. 2.5. The coordinates of the two masses, x and y, with the reference directions as shown, are the displacements of the masses from their equilibrium conditions. The equilibrium positions are offset from the springs’ unstretched positions because of the force of gravity. The shock absorber is represented in the schematic diagram by a dashpot symbol with friction constant b. The magnitude of the force from the shock

Figure 2.4 Automobile suspension

Figure 2.5 The quarter-car model

y

m2 ks

ycar

b x

m1 kw

Road surface r

Inertial reference

26 Chapter 2 Dynamic Models Figure 2.6 Free-body diagrams for suspension system

b(y
x)

ks(y
x)

x

y

m1

m2

kw(x
r)

b(y
x)

ks(y
x)

absorber is assumed to be proportional to the rate of change of the relative displacement of the two masses—that is, the force = b(˙y − x˙ ). The force of gravity could be included in the free-body diagram; however, its effect is to produce a constant offset of x and y. By defining x and y to be the distance from the equilibrium position, the need to include the gravity forces is eliminated. The force from the car suspension acts on both masses in proportion to their relative displacement with spring constant ks . Figure 2.6 shows the free-body diagram of each mass. Note that the forces from the spring on the two masses are equal in magnitude but act in opposite directions, which is also the case for the damper. A positive displacement y of mass m2 will result in a force from the spring on m2 in the direction shown and a force from the spring on m1 in the direction shown. However, a positive displacement x of mass m1 will result in a force from the spring ks on m1 in the opposite direction to that drawn in Fig. 2.6, as indicated by the minus x term for the spring force. The lower spring kw represents the tire compressibility, for which there is insufficient damping (velocity-dependent force) to warrant including a dashpot in the model. The force from this spring is proportional to the distance the tire is compressed and the nominal equilibrium force would be that required to support m1 and m2 against gravity. By defining x to be the distance from equilibrium, a force will result if either the road surface has a bump (r changes from its equilibrium value of zero) or the wheel bounces (x changes). The motion of the simplified car over a bumpy road will result in a value of r(t) that is not constant. As previously noted, there is a constant force of gravity acting on each mass; however, this force has been omitted, as have the equal and opposite forces from the springs. Gravitational forces can always be omitted from vertical-spring mass systems (1) if the position coordinates are defined from the equilibrium position that results when gravity is acting, and (2) if the spring forces used in the analysis are actually the perturbation in spring forces from those forces acting at equilibrium. Applying Eq. (2.1) to each mass and noting that some forces on each mass are in the negative (down) direction yields the system of equations b(˙y − x˙ ) + ks (y − x) − kw (x − r) = m1 x¨ ,

(2.8)

−ks (y − x) − b(˙y − x˙ ) = m2 y¨ .

(2.9)

Some rearranging results in b x¨ + (˙x − y˙ ) + m1 b y¨ + (˙y − x˙ ) + m2

ks kw kw (x − y) + x= r, m1 m1 m1 ks (y − x) = 0. m2

(2.10) (2.11)

2.1 Dynamics of Mechanical Systems

Check for sign errors

27

The most common source of error in writing equations for systems like these are sign errors. The method for keeping the signs straight in the preceding development entailed mentally picturing the displacement of the masses and drawing the resulting force in the direction that the displacement would produce. Once you have obtained the equations for a system, a check on the signs for systems that are obviously stable from physical reasoning can be quickly carried out. As we will see when we study stability in Section 3.6, a stable system always has the same signs on similar variables. For this system, Eq. (2.10) shows that the signs on the x¨ , x˙ , and x terms are all positive, as they must be for stability. Likewise, the signs on the y¨ , y˙ , and y terms are all positive in Eq. (2.11). The transfer function is obtained in a similar manner as before. Substituting s for d/dt in the differential equations yields s2 X(s) + s

b ks kw kw (X(s) − Y (s)) + (X(s) − Y (s)) + X(s) = R(s), m1 m1 m1 m1 b ks s2 Y (s) + s (Y (s) − X(s)) + (Y (s) − X(s)) = 0, m2 m2

which, after some algebra and rearranging, yields the transfer function   ks kw b s+ Y (s) m m b   1 2     . = b k kw b b k k k w ks R (s) s s w 4 3 2 s + + + + s + s + s+ m1 m2 m1 m2 m1 m1 m2 m1 m2 (2.12) To determine numerical values, we subtract the mass of the four wheels from the total car mass of 1580 kg and divide by 4 to find that m2 = 375 kg. The wheel mass was measured directly to be m1 = 20 kg. Therefore, the transfer function with the numerical values is Y (s) 1.31e06(s + 13.3) = 4 . (2.13) R(s) s + (516.1)s3 + (5.685e04)s2 + (1.307e06)s + 1.733e07

2.1.2 Rotational Motion Newton’s law for rotational motion

Application of Newton’s law to one-dimensional rotational systems requires that Eq. (2.1) be modified to M = Iα, (2.14) where M = the sum of all external moments about the center of mass of a body, N·m or lb·ft, I = the body’s mass moment of inertia about its center of mass, kg·m2 or slug·ft2 , α = the angular acceleration of the body, rad/sec2 .

28 Chapter 2 Dynamic Models

EXAMPLE 2.3

Rotational Motion: Satellite Attitude Control Model Satellites, as shown in Fig. 2.7, usually require attitude control so that antennas, sensors, and solar panels are properly oriented. Antennas are usually pointed toward a particular location on earth, while solar panels need to be oriented toward the sun for maximum power generation. To gain insight into the full three-axis attitude control system, it is helpful to consider one axis at a time. Write the equations of motion for one axis of this system and show how they would be depicted in a block diagram. In addition, determine the transfer function of this system and construct the system as if it were to be evaluated via MATLAB’s SIMULINK. Solution. Figure 2.8 depicts this case, where motion is allowed only about the axis perpendicular to the page. The angle θ that describes the satellite orientation must be measured with respect to an inertial reference—that is, a reference that has no angular acceleration. The control force comes from reaction jets that produce a moment of Fc d about the mass center. There may also be small disturbance moments MD on the

Figure 2.7 Communications satellite Source: Courtesy Space Systems/Loral

2.1 Dynamics of Mechanical Systems Figure 2.8 Satellite control schematic

29

u u

MD

d Gas jet Fc Inertial reference

Figure 2.9 Block diagrams representing Eq. (2.15) in the upper half and Eq. (2.16) in the lower half

Fc d + MD

1 I

U(s)

1 I

 ␪

s2 (s)

∫ 1 s

 ␪

s (s)

␪

∫ 1 s

(s)

satellite, which arise primarily from solar pressure acting on any asymmetry in the solar panels. Applying Eq. (2.14) yields the equation of motion ¨ Fc d + MD = I θ. Double-integrator plant

The output of this system, θ, results from integrating the sum of the input torques twice; hence this type of system is often referred to as the double-integrator plant. The transfer function can be obtained as described for Eq. (2.7) and is (s) 1 1 = , U(s) I s2

1/s2 plant

(2.15)

(2.16)

where U = Fc d + MD. In this form, the system is often referred to as the 1/s2 plant. Figure 2.9 shows a block diagram representing Eq. (2.15) in the upper half and a block diagram representing Eq. (2.16) in the lower half. This simple system can be analyzed using the linear analysis techniques that are described in later chapters, or via MATLAB as we saw in Example 2.1. It can also be numerically evaluated for an arbitrary input time history using SIMULINK. SIMULINK is a sister software package to MATLAB for interactive, nonlinear simulation and has a graphical user interface with drag and drop properties. Figure 2.10 shows a block diagram of the system as depicted by SIMULINK.

30 Chapter 2 Dynamic Models Figure 2.10 SIMULINK block diagram of the double-integrator plant

1

1

U

1/I

1 s

1 s

Integrator 1 Integrator 2

Scope

Figure 2.11 Disk read/write mechanism Source: Courtesy of Hewlett-Packard Company

In many cases a system, such as the disk-drive read/write head shown in Fig. 2.11, in reality has some flexibility, which can cause problems in the design of a control system. Particular difficulty arises when there is flexibility, as in this case, between the sensor and actuator locations. Therefore, it is often important to include this flexibility in the model even when the system seems to be quite rigid.

EXAMPLE 2.4

Flexibility: Flexible Read/Write for a Disk Drive Assume that there is some flexibility between the read head and the drive motor in Fig. 2.11. Find the equations of motion relating the motion of the read head to a torque applied to the base.

2.1 Dynamics of Mechanical Systems Figure 2.12 Disk read/write head schematic for modeling

Read head and track sensor

Disk

31

Head inertia I2

u2

Flexible shaft k, b u1

Motor inertia I1

Mc + MD

Figure 2.13 Free-body diagrams of the disk read/write head

u2

u1 u1

u2 k(u1
u2) k(u1
u2)

Mc  M D

b(u1
u2)

I1

b(u1
u2)

I2

Solution. The dynamic model for this situation is shown schematically in Fig. 2.12. This model is dynamically similar to the resonant system shown in Fig. 2.5 and results in equations of motion that are similar in form to Eqs. (2.10) and (2.11). The moments on each body are shown in the free-body diagrams in Fig. 2.13. The discussion of the moments on each body is essentially the same as the discussion for Example 2.2, except that the springs and damper in that case produced forces, instead of moments that act on each inertia, as in this case. When the moments are summed, equated to the accelerations according to Eq. (2.14), and rearranged, the result is I1 θ¨1 + b(θ˙1 − θ˙2 ) + k(θ1 − θ2 ) = Mc + MD ,

(2.17)

I2 θ¨2 + b(θ˙2 − θ˙1 ) + k(θ2 − θ1 ) = 0.

(2.18)

Ignoring the disturbance torque MD and the damping b for simplicity, we find the transfer function from the applied torque Mc to the read head motion to be 2 (s) = Mc (s)

k

 I1 I2 s

2

k k s + + I1 I2

.

(2.19)

2

It might also be possible to sense the motion of the inertia where the torque is applied, θ1 , in which case the transfer function with the same simplifications would be 1 (s) = Mc (s)

I2 s 2 + k  . k k I1 I2 s2 s2 + + I1 I2

(2.20)

32 Chapter 2 Dynamic Models Figure 2.14 Pendulum Tc l u

mg

Collocated sensor and actuator

These two cases are typical of many situations in which the sensor and actuator may or may not be placed in the same location in a flexible body. We refer to the situation between sensor and actuator in Eq. (2.19) as the “noncollocated” case, whereas Eq. (2.20) describes the “collocated” case. You will see in Chapter 5 that it is far more difficult to control a system when there is flexibility between the sensor and actuator (noncollocated case) than when the sensor and actuator are rigidly attached to one another (the collocated case). In the special case in which a point in a rotating body is fixed with respect to an inertial reference, as is the case with a pendulum, Eq. (2.14) can be applied such that M is the sum of all moments about the fixed point and I is the moment of inertia about the fixed point.

EXAMPLE 2.5

Rotational Motion: Pendulum 1. Write the equations of motion for the simple pendulum shown in Fig. 2.14, where all the mass is concentrated at the end point and there is a torque, Tc , applied at the pivot. 2. Use MATLAB to determine the time history of θ to a step input in Tc of 1 N·m. Assume l = 1 m, m = 1 kg, and g = 9.81 m/sec2 . Solution 1. Equations of motion: The moment of inertia about the pivot point is I = ml 2 . The sum of moments about the pivot point contains a term from gravity as well as the applied torque Tc . The equation of motion, obtained from Eq. (2.14), is ¨ (2.21) Tc − mgl sin θ = I θ, which is usually written in the form Tc g θ¨ + sin θ = 2 . (2.22) l ml This equation is nonlinear due to the sin θ term. A general discussion of nonlinear equations is contained in Chapter 9; however, we can proceed with a linearization of this case by assuming the motion is small enough that sin θ ∼ = θ. Then Eq. (2.22) becomes the linear equation Tc g θ¨ + θ = 2 . (2.23) l ml

2.1 Dynamics of Mechanical Systems

33

With no applied torque, the natural motion is that of a harmonic oscillator with a natural frequency of 3  g . (2.24) ωn = l The transfer function can be obtained as described for Eq. (2.7), yielding 1 (s) 2 = ml g . (2.25) Tc (s) s2 + l 2. Time history: The dynamics of a system can be prescribed to MATLAB in terms of row vectors containing the coefficients of the polynomials describing the numerator and denominator of its transfer function. In this case, the numerator (called num) is simply one number, since there are no powers of s, so that 1 1 num = 2 = = [1], ml (1)(1)2 and the denominator (called den) contains the coefficients of the descending powers of s in (s2 + g/l) and is a row vector with three elements:  g = [1 0 9.81]. den = 1 0 l The desired response of the system can be obtained by using the MATLAB step response function, called step. The MATLAB statements t = 0:0.02:10; num = 1; den = [1 0 9.81]; sys = tf(num,den); y = step(sys,t); plot(t, 57.3*y)

% vector of times for output, 0 to 10 at 0.02 increments

% defines the system by its numerator and denominator % computes step responses at times given by t for step at t = 0 % converts radians to degrees and plots step response

will produce the desired time history shown in Fig. 2.15.

SIMULINK

As we saw in this example, the resulting equations of motion are often nonlinear. Such equations are much more difficult to solve than linear ones, and the kinds of possible motions resulting from a nonlinear model are much more difficult to categorize than those resulting from a linear model. It is therefore useful to linearize models in order to gain access to linear analysis methods. It may be that the linear models and linear analysis are used only for the design of the control system (whose function may be to maintain the system in the linear region). Once a control system is synthesized and shown to have desirable performance based on linear analysis, it is then prudent to carry out further analysis or an accurate numerical simulation of the system with the significant nonlinearities in order to validate that performance. SIMULINK is an expedient way to carry out these simulations and can handle most 3 In a grandfather clock it is desired to have a pendulum period of exactly 2 sec. Show that the pendulum should be approximately 1 m in length.

34 Chapter 2 Dynamic Models 12

10 Pendulum angle, u (°)

Figure 2.15 Response of the pendulum to a step input of 1 N-m in the applied torque

8

6

4

2

0

0

2

4

6

8

10

Time (sec)

Figure 2.16 The SIMULINK block diagram representing the linear equation (2.26)

1 Step

Gain 1

Ts

ⴙ ⴚ

 

1 s Integrator 9.81* 

 

 1 s Integrator 1

57.3 Gain 2

 (°) Scope

9.81 Gain

nonlinearities. Use of this simulation tool is carried out by constructing a block diagram4 that represents the equations of motion. The linear equation of motion for the pendulum with the parameters as specified in Example 2.5 can be seen from Eq. (2.23) to be θ¨ = −9.81 ∗ θ + 1,

(2.26)

and this is represented in SIMULINK by the block diagram in Fig. 2.16. Note that the circle on the left side of the figure with the + and − signs indicating addition and subtraction implements the equation above. The result of running this numerical simulation will be essentially identical to the linear solution shown in Fig. 2.15 because the solution is for relatively small angles where sin θ ∼ = θ . However, using SIMULINK to solve for the response enables us to simulate the nonlinear equation so that we could analyze the system for larger motions. In this case, Eq. (2.26) becomes θ¨ = −9.81 ∗ sin θ + 1,

4A more extensive discussion of block diagrams is contained in Section 3.2.1.

(2.27)

2.1 Dynamics of Mechanical Systems Figure 2.17 The SIMULINK block diagram representing the nonlinear equation (2.27)

Step

1 S

ⴙ ⴚ

1

1 S

Gain 1

9.81 Gain 3

Figure 2.18 Block diagram of the pendulum for both the linear and nonlinear models

K Step

ⴙ ⴚ

Gain 1

1 s Integrator

57.3

Integrator 2

Integrator 3

35

Scope

sin Trigonometric function

1 s Integrator 1 9.81 Gain

ⴙ ⴚ

1 s Integrator 3 9.81 Gain 3

1 s Integrator 2

57.3 Mux

Gain 2

Scope

sin Trigonometric function

and the SIMULINK block diagram shown in Fig. 2.17 implements this nonlinear equation. SIMULINK is capable of simulating all commonly encountered nonlinearities, including deadzones, on–off functions, stiction, hysteresis, aerodynamic drag (a function of v2 ), and trigonometric functions. All real systems have one or more of these characteristics in varying degrees.

EXAMPLE 2.6

Use of SIMULINK for Nonlinear Motion: Pendulum Use SIMULINK to determine the time history of θ for the pendulum in Example 2.5. Compare it against the linear solution for Tc values of 1 N·m and 4 N·m. Solution. Time history: The SIMULINK block diagrams for the two cases discussed above are combined and both outputs in Fig. 2.16 and 2.17 are sent via a “multiplexer block (Mux)” to the “scope” so they can be plotted on the same graph. Fig. 2.18 shows the combined block diagram where the gain, K, represents the values of Tc . The outputs of this system for Tc values of 1 N · m and 4 N· m are shown in Fig. 2.19. Note that for Tc = 1 N·m, the outputs at the top of the figure remain at 12◦ or less and the linear approximation is extremely close to the nonlinear output. For Tc = 4 N·m, the output angle grows to near 50◦ and a substantial difference in the response magnitude and frequency is apparent due to θ being a poor approximation to sin θ at these magnitudes.

36 Chapter 2 Dynamic Models (a) Tc = 1 N.m

50 45 40 35

u (°)

30 25 20 Linear response

15

Nonlinear response

10 5 0 (b) Tc = 4 N.m

50

Nonlinear

Linear

45 40 35 30 u (°)

Figure 2.19 Response of the pendulum SIMULINK numerical simulation for the linear and nonlinear models. (a) for Tc = 1 N·m and (b) Tc = 4 N·m

25 20 15 10 5 0

0

1

2

3

4

5 Time (sec)

6

7

8

9

10

Chapter 9 is devoted to the analysis of nonlinear systems and greatly expands on these ideas.

2.1.3 Combined Rotation and Translation In some cases, mechanical systems contain both translational and rotational portions. The procedure is the same as that described in Sections 2.1.1 and 2.1.2: sketch the free-body diagrams, define coordinates and positive directions, determine all forces and moments acting, and apply Eqs. (2.1) and/or (2.14). An exact derivation of the equations for these systems can become quite involved; therefore, the complete analysis for the following examples are contained in Appendix W2 and only the linearized equations of motion and their transfer functions are given here.

2.1 Dynamics of Mechanical Systems Figure 2.20 Schematic of the crane with hanging load

37

x

mt

u

I, mp u

Figure 2.21 Inverted pendulum

u

I, mp

u

mt

x

EXAMPLE 2.7

Rotational and Translational Motion: Hanging Crane Write the equations of motion for the hanging crane shown schematically in Fig. 2.20. Linearize the equations about θ = 0, which would typically be valid for the hanging crane. Also linearize the equations for θ = π, which represents the situation for the inverted pendulum shown in Fig. 2.21. The trolley has mass, mt , and the hanging crane (or pendulum) has mass, mp , and inertia about its mass center of I. The distance from the pivot to the mass center of the pendulum is l; therefore, the moment of inertia of the pendulum about the pivot point is (I + mp l 2 ). Solution. Free-body diagrams need to be drawn for the trolley and the pendulum and the reaction forces considered where the two attach to one another. We carry out this process in Appendix W2. After Newton’s Laws are applied for the translational motion of the trolley and the rotational motion of the pendulum, it will be found that the reaction forces between the two bodies can be eliminated, and the only unknowns will be θ and x. The results are two coupled second-order nonlinear differential equations in θ and x with the input being the force applied to the trolley, u. They can be linearized in a similar manner that was done for the simple pendulum by assuming small angles. For small motions about θ = 0, we let cos θ ∼ = θ , and = 1, sin θ ∼ θ˙ 2 ∼ = 0; thus the equations are approximated by (I + mp l2 )θ¨ + mp glθ = −mp l¨x , (mt + mp )¨x + b˙x + mp lθ¨ = u.

(2.28)

Note that the first equation is very similar to the simple pendulum, Eq. (2.21), where the applied torque arises from the trolley accelerations. Likewise, the second equation representing the trolley motion, x, is very similar to the car translation in

38 Chapter 2 Dynamic Models Eq. (2.3) where the forcing term arises from the angular acceleration of the pendulum. Neglecting the friction term b leads to the transfer function from the control input u to hanging crane angle θ : −mp l θ (s) = . 2 U(s) ((I + mp l )(mt + mp ) − mp2 l2 )s2 + mp gl(mt + mp )

(2.29)

For the inverted pendulum in Fig. 2.21, where θ ∼ = π, assume θ = π + θ
, where represents motion from the vertical upward direction. In this case, sin θ ∼ = −θ
, 5 ∼ cos θ = −1, and the nonlinear equations become

θ
Inverted pendulum equations

(I + mp l2 )θ¨
− mp glθ
= mp l¨x , (mt + mp )¨x + b˙x − mp lθ¨
= u.

(2.30)

As noted in Example 2.2, a stable system will always have the same signs on each variable, which is the case for the stable hanging crane modeled by Eqs. (2.28). However, the signs on θ and θ¨ in the top Eq. (2.30) are opposite, thus indicating instability, which is the characteristic of the inverted pendulum. The transfer function, again without friction, is mp l θ
(s) = . 2 2 2 U(s) ((I + mp l ) − mp l )s2 − mp gl(mt + mp )

(2.31)

In Chapter 5 you will learn how to stabilize systems using feedback and will see that even unstable systems like an inverted pendulum can be stabilized providing there is a sensor that measures the output quantity and a control input. For the case of the inverted pendulum perched on a trolley, it would be required to measure the pendulum angle, θ
, and provide a control input, u, that accelerated the trolley in such a way that the pendulum remained pointing straight up. In years past, this system existed primarily in university control system laboratories as an educational tool. However, more recently, there is a practical device in production and being sold that employs essentially this same dynamic system: The Segway. It uses a gyroscope so that the angle of the device is known with respect to vertical, and electric motors provide a torque on the wheels so that it balances the device and provides the desired forward or backward motion. It is shown in Fig. 2.22.

2.1.4 Distributed Parameter Systems All the preceding examples contained one or more rigid bodies, although some were connected to others by springs. Actual structures—for example, satellite solar panels, airplane wings, or robot arms—usually bend, as shown by the flexible beam in Fig. 2.23(a). The equation describing its motion is a fourth-order partial differential equation that arises because the mass elements are continuously distributed along the beam with a small amount of flexibility between elements. This type of system is called a distributed parameter system. The dynamic analysis methods presented 5 The inverted pendulum is often described with the angle of the pendulum being positive for clockwise motion. If defined that way, then reverse the sign on all terms in Eqs. (2.30) in θ
or θ¨
.

2.1 Dynamics of Mechanical Systems

39

Figure 2.22 The Segway, which is similar to the inverted pendulum and is kept upright by a feedback control system Source: Photo courtesy of David Powell

in this section are not sufficient to analyze this case; however, more advanced texts (Thomson and Dahleh, 1998) show that the result is EI

∂ 4w ∂ 2w + ρ = 0, ∂x 4 ∂t 2

(2.32)

where E = Young’s modulus, I = beam area moment of inertia, ρ = beam density, w = beam deflection at length x along the beam. The exact solution to Eq. (2.32) is too cumbersome to use in designing control systems, but it is often important to account for the gross effects of bending in control systems design. The continuous beam in Fig. 2.23(b) has an infinite number of vibration-mode shapes, all with different frequencies. Typically, the lowest-frequency modes have the largest amplitude and are the most important to approximate well. The simplified model in Fig. 2.23(c) can be made to duplicate the essential behavior of the first

40 Chapter 2 Dynamic Models Figure 2.23 (a) Flexible robot arm used for research at Stanford University; (b) model for a continuous flexible beam; (c) simplified model for the first bending mode; (d) model for the first and second bending modes Source: Photo courtesy of E. Schmitz

w

(b)

A flexible structure can be approximated by a lumped parameter model

(c)

(d)

bending mode shape and frequency and would usually be adequate for controller design. If frequencies higher than the first bending mode are anticipated in the control system operation, it may be necessary to model the beam as shown in Fig. 2.23(d), which can be made to approximate the first two bending modes and frequencies. Likewise, higher-order models can be used if such accuracy and complexity are deemed necessary (Thomson and Dahleh, 1998; Schmitz, 1985). When a continuously bending object is approximated as two or more rigid bodies connected by springs, the resulting model is sometimes referred to as a lumped parameter model.

2.1.5 Summary: Developing Equations of Motion for Rigid Bodies The physics necessary to write the equations of motion of a rigid body is entirely given by Newton’s laws of motion. The method is as follows: 1. Assign variables such as x and θ that are both necessary and sufficient to describe an arbitrary position of the object. 2. Draw a free-body diagram of each component. Indicate all forces acting on each body and their reference directions. Also indicate the accelerations of the center of mass with respect to an inertial reference for each body.

2.2 Models of Electric Circuits

41

3. Apply Newton’s law in translation [Eq. (2.1)] and/or rotation [Eq. (2.14)] form. 4. Combine the equations to eliminate internal forces. 5. The number of independent equations should equal the number of unknowns.

2.2 Models of Electric Circuits

Kirchhoff’s laws

Electric circuits are frequently used in control systems largely because of the ease of manipulation and processing of electric signals. Although controllers are increasingly implemented with digital logic, many functions are still performed with analog circuits. Analog circuits are faster than digital and, for very simple controllers, an analog circuit would be less expensive than a digital implementation. Furthermore, the power amplifier for electromechanical control and the anti-alias prefilters for digital control must be analog circuits. Electric circuits consist of interconnections of sources of electric voltage and current, and other electronic elements such as resistors, capacitors, and transistors. An important building block for circuits is an operational amplifier (or op-amp),6 which is also an example of a complex feedback system. Some of the most important methods of feedback system design were developed by the designers of high-gain, widebandwidth feedback amplifiers, mainly at the Bell Telephone Laboratories between 1925 and 1940. Electric and electronic components also play a central role in electromechanical energy conversion devices such as electric motors, generators, and electrical sensors. In this brief survey we cannot derive the physics of electricity or give a comprehensive review of all the important analysis techniques. We will define the variables, describe the relations imposed on them by typical elements and circuits, and describe a few of the most effective methods available for solving the resulting equations. Symbols for some linear circuit elements and their current–voltage relations are given in Fig. 2.24. Passive circuits consist of interconnections of resistors, capacitors, and inductors. With electronics, we increase the set of electrical elements by adding active devices, including diodes, transistors, and amplifiers. The basic equations of electric circuits, called Kirchhoff’s laws, are as follows: 1. Kirchhoff’s current law (KCL): The algebraic sum of currents leaving a junction or node equals the algebraic sum of currents entering that node. 2. Kirchhoff’s voltage law (KVL): The algebraic sum of all voltages taken around a closed path in a circuit is zero. With complex circuits of many elements, it is essential to write the equations in a careful, well organized way. Of the numerous methods for doing this, we choose for description and illustration the popular and powerful scheme known as node analysis. One node is selected as a reference and we assume the voltages of all other nodes to be unknowns. The choice of reference is arbitrary in theory, but in actual electronic 6 Oliver Heaviside introduced the mathematical operation p to signify differentiation so that pv = dv/dt. The

Laplace transform incorporates this idea, using the complex variable s. Ragazzini et al. (1947) demonstrated that an ideal, high-gain electronic amplifier permitted one to realize arbitrary “operations” in the Laplace transform variable s, so they named it the operational amplifier, commonly abbreviated to op-amp.

42 Chapter 2 Dynamic Models Figure 2.24 Elements of electric circuits

Symbol

Equation

 Resistor

i

y

y  Ri


 Capacitor

y

i

dy i  C dt

i

di y  L dt

ys

y  ys


 Inductor

y


Voltage source

y

ⵑ


i Current source

i  is is

circuits the common, or ground, terminal is the obvious and standard choice. Next, we write equations for the selected unknowns using the current law (KCL) at each node. We express these currents in terms of the selected unknowns by using the element equations in Fig. 2.24. If the circuit contains voltage sources, we must substitute a voltage law (KVL) for such sources. Example 2.8 illustrates how node analysis works.

EXAMPLE 2.8

Equations for the Bridged Tee Circuit Determine the differential equations for the circuit shown in Fig. 2.25. Solution. We select node 4 as the reference and the voltages v1 , v2 , and v3 at nodes 1, 2, and 3 as the unknowns. We start with the degenerate KVL relationship v1 = vi .

(2.33)

At node 2 the KCL is −

v1 − v2 dv2 v2 − v 3 = 0, + + C1 R1 R2 dt

(2.34)

2.2 Models of Electric Circuits Figure 2.25 Bridged tee circuit

43

C2 
 R1
2  R2


1  yi

ⵑ

y1







C1

3 

y2 y3

y0


4

and at node 3 the KCL is

v3 − v 2 d(v3 − v1 ) = 0. + C2 R2 dt These three equations describe the circuit.

Operational amplifier

Kirchhoff’s laws can also be applied to circuits that contain an operational amplifier. The simplified circuit of the op-amp is shown in Fig. 2.26(a) and the schematic symbol is drawn in Fig. 2.26(b). If the positive terminal is not shown, it is assumed to be connected to ground, v+ = 0, and the reduced symbol of Fig. 2.26(c) is used. For use in control circuits, it is usually assumed that the op-amp is ideal with the values R1 = ∞, R0 = 0, and A = ∞. The equations of the ideal op-amp are extremely simple, being (2.36) i+ = i− = 0,

vout Figure 2.26 (a) Op-amp simplified circuit; (b) op-amp schematic symbol; (c) reduced symbol for v+ = 0

(2.35)

(2.37) v+ − v− = 0. The gain of the amplifier is assumed to be so high that the output voltage becomes = whatever it takes to satisfy these equations. Of course, a real amplifier only i


R0

y


R1

i

y

i0

苲

y0  A(y
y
)




(a) y
y

i
i




(b)

y0

y





(c)

y0

44 Chapter 2 Dynamic Models Figure 2.27 The op-amp summer

Rf

y1

i1

R1

iout

yout




y2 i2

R2

approximates these equations, but unless they are specifically described, we will assume all op-amps are ideal. More realistic models are the subject of several problems given at the end of the chapter.

EXAMPLE 2.9

Op-Amp Summer Find the equations and transfer functions of the circuit shown in Fig. 2.27. Solution. Equation (2.37) requires that v− = 0, and thus the currents are i1 = v1/R1 , i2 = v2/R2 , and iout = vout/Rf . To satisfy Eq. (2.36), i1 + i2 + iout = 0, from which it follows that v1/R1 + v2/R2 + vout/Rf = 0, and we have   Rf Rf vout = − v1 + v2 . (2.38) R1 R2

The op-amp summer

From this equation we see that the circuit output is a weighted sum of the input voltages with a sign change. The circuit is called a summer. A second important example for control is given by the op-amp integrator.

EXAMPLE 2.10

Integrator

Op-amp as integrator

Find the transfer function for the circuit shown in Fig. 2.28. Solution. In this case the equations are differential and Eqs. (2.36) and (2.37) require iin + iout = 0,

(2.39)

vin dvout = 0. +C Rin dt

(2.40)

so that

Figure 2.28 The op-amp integrator

C

yin


Rin

iout

yout

2.3 Models of Electromechanical Systems Eq. (2.40) can be written in integral form as

t 1 vin (τ ) dτ + vout (0). vout = − Rin C 0

45

(2.41)

Using the operational notation that d/dt = s in Eq. (2.40), the transfer function (which assumes zero initial conditions) can be written as Vout (s) = −

1 Vin (s) . s Rin C

(2.42)

Thus the ideal op-amp in this circuit performs the operation of integration and the circuit is simply referred to as an integrator.

2.3 Models of Electromechanical Systems Electric current and magnetic fields interact in two ways that are particularly important to an understanding of the operation of most electromechanical actuators and sensors. If a current of i amperes in a conductor of length l meters is arranged at right angles in a magnetic field of B tesls, then there is a force on the conductor at right angles to the plane of i and B, with magnitude F = Bli newtons.

Law of motors

(2.43)

This equation is the basis of conversion of electric energy to mechanical work and is called the law of motors.

EXAMPLE 2.11

Modeling a Loudspeaker A typical geometry for a loudspeaker for producing sound is sketched in Fig. 2.29. The permanent magnet establishes a radial field in the cylindrical gap between the poles of the magnet. The force on the conductor wound on the bobbin causes the

Figure 2.29 Geometry of a loudspeaker: (a) overall configuration; (b) the electromagnet and voice coil

Cone Cone

S

Electromagnet N Coil Low-force suspension Permanent magnet (a)

(b)

S

Bobbin

46 Chapter 2 Dynamic Models voice coil to move, producing sound.7 The effects of the air can be modeled as if the cone had equivalent mass M and viscous friction coefficient b. Assume that the magnet establishes a uniform field B of 0.5 tesla and the bobbin has 20 turns at a 2-cm diameter. Write the equations of motion of the device. Solution. The current is at right angles to the field, and the force of interest is at right angles to the plane of i and B, so Eq. (2.43) applies. In this case the field strength is B = 0.5 tesla and the conductor length is 2π l = 20 × m = 1.26 m. 100 Thus, the force is F = 0.5 × 1.26 × i = 0.63i N. The mechanical equation follows from Newton’s laws, and for a mass M and friction coefficient b, the equation is M x¨ + b˙x = 0.63i.

(2.44)

This second-order differential equation describes the motion of the loudspeaker cone as a function of the input current i driving the system. Substituting s for d/dt in Eq. (2.44) as before, the transfer function is easily found to be X(s) 0.63/M = . (2.45) I(s) s(s + b/M) The second important electromechanical relationship is the effect of mechanical motion on electric voltage. If a conductor of length l meters is moving in a magnetic field of B teslas at a velocity of v meters per second at mutually right angles, an electric voltage is established across the conductor with magnitude e(t) = Blv V.

Law of generators

(2.46)

This expression is called the law of generators.

EXAMPLE 2.12

Loudspeaker with Circuit For the loudspeaker in Fig. 2.29 and the circuit driving it in Fig. 2.30, find the differential equations relating the input voltage va to the output cone displacement x. Assume the effective circuit resistance is R and the inductance is L. Solution. The loudspeaker motion satisfies Eq. (2.44), and the motion results in a voltage across the coil as given by Eq. (2.46), with the velocity x˙ . The resulting voltage is (2.47) ecoil = Blx˙ = 0.63˙x . This induced voltage effect needs to be added to the analysis of the circuit. The equation of motion for the electric circuit is di (2.48) L + Ri = va − 0.63˙x . dt 7 Similar voice-coil motors are commonly used as the actuator for the read/write head assembly of computer

hard-disk data access devices.

2.3 Models of Electromechanical Systems Figure 2.30 A loudspeaker showing the electric circuit

i

R

 ya

47

x

L

ⵑ


 ecoil

ⵑ


These two coupled equations, (2.44) and (2.48), constitute the dynamic model for the loudspeaker. Again substituting s for d/dt in these equations, the transfer function between the applied voltage and the loudspeaker displacement is found to be 0.63 X(s) . =  Va (s) s (Ms + b)(Ls + R) + (0.63)2

DC motor actuators

Back emf

(2.49)

A common actuator based on these principles and used in control systems is the DC motor to provide rotary motion. A sketch of the basic components of a DC motor is given in Fig. 2.31. In addition to housing and bearings, the nonturning part (stator) has magnets, which establish a field across the rotor. The magnets may be electromagnets or, for small motors, permanent magnets. The brushes contact the rotating commutator, which causes the current always to be in the proper conductor windings so as to produce maximum torque. If the direction of the current is reversed, the direction of the torque is reversed. The motor equations give the torque T on the rotor in terms of the armature current ia and express the back emf voltage in terms of the shaft’s rotational velocity θ˙m .8

Figure 2.31 Sketch of a DC motor

N

Stator magnet

Brush

ia

Rotor windings

Shaft angle um

Stator magnet S

Shaft Bearings

Brush Commutator

8 Because the generated electromotive force (emf) works against the applied armature voltage, it is called the back emf.

48 Chapter 2 Dynamic Models Thus T = K t ia , e = Ke θ˙m .

(2.50) (2.51)

Torque

In consistent units, the torque constant Kt equals the electric constant Ke , but in some cases the torque constant will be given in other units, such as ounce-inches per ampere, and the electric constant may be expressed in units of volts per 1000 rpm. In such cases the engineer must make the necessary translations to be certain the equations are correct.

EXAMPLE 2.13

Modeling a DC Motor Find the equations for a DC motor with the equivalent electric circuit shown in Fig. 2.32(a). Assume that the rotor has inertia Jm and viscous friction coefficient b. Solution. The free-body diagram for the rotor, shown in Fig. 2.32(b), defines the positive direction and shows the two applied torques, T and bθ˙m . Application of Newton’s laws yields Jm θ¨m + bθ˙m = Kt ia . (2.52) Analysis of the electric circuit, including the back emf voltage, shows the electrical equation to be dia + Ra ia = va − Ke θ˙m . La (2.53) dt With s substituted for d/dt in Eqs. (2.52) and (2.53), the transfer function for the motor is readily found to be Kt m (s) = . Va (s) s[(Jm s + b)(La s + Ra ) + Kt Ke ]

(2.54)

In many cases the relative effect of the inductance is negligible compared with the mechanical motion and can be neglected in Eq. (2.53). If so, we can combine Eqs. (2.52) and (2.53) into one equation to get   K t Ke Kt ¨ va . (2.55) Jm θm + b + θ˙m = Ra Ra Figure 2.32 DC motor: (a) electric circuit of the armature; (b) free-body diagram of the rotor

T Ra

La ia

 ya

ⵑ

um 

e  Keum





Jm bum (a)

(b)

2.3 Models of Electromechanical Systems

49

From Eq. (2.55) it is clear that in this case the effect of the back emf is indistinguishable from the friction, and the transfer function is Kt m (s) Ra   (2.56) = Kt Ke Va (s) 2 s Jm s + b + Ra K = , (2.57) s(τ s + 1) where Kt K= , (2.58) bRa + Kt Ke Ra Jm τ= . (2.59) bRa + Kt Ke In many cases, a transfer function between the motor input and the output speed (ω = θ˙m ) is required. In such cases, the transfer function would be (s) m (s) K =s = . Va (s) Va (s) τs + 1

(2.60)

Another device used for electromechanical energy conversion is the alternating current (AC) induction motor invented by N. Tesla. Elementary analysis of the AC motor is more complex than that of the DC motor. A typical experimental set of curves of torque versus speed for fixed frequency and varying amplitude of applied (sinusoidal) voltage is given in Fig. 2.33. Although the data in the figure are for a constant engine speed, they can be used to extract the motor constants that will provide a dynamic model for the motor. For analysis of a control problem involving an AC

AC motor actuators

ya  V4

ya  V2 ( V1) Torque, T

Torque, T

V3

y a  V1

V2

Slope K2

V1

Speed, um (a)

Speed, um (b)

Figure 2.33 Torque–speed curves for a servo motor showing four amplitudes of armature voltage: (a) low-rotor-resistance machine; (b) high-rotor-resistance machine showing four values of armature voltage, va

50 Chapter 2 Dynamic Models motor such as that described by Fig. 2.33, we make a linear approximation to the curves for speed near zero and at a midrange voltage to obtain the expression T = K1 va − K2 θ˙m .

(2.61)

The constant K1 represents the ratio of a change in torque to a change in voltage at zero speed and is proportional to the distance between the curves at zero speed. The constant K2 represents the ratio of a change in torque to a change in speed at zero speed and a midrange voltage; therefore, it is the slope of a curve at zero speed as shown by the line at V2 . For the electrical portion, values for the armature resistance Ra and inductance La are also determined by experiment. Once we have values for K1 , K2 , Ra , and La , the analysis proceeds as the analysis in Example 2.13 for the DC motor. For the case in which the inductor can be neglected, we can substitute K1 and K2 into Eq. (2.55) in place of Kt /Ra and Kt Ke /Ra , respectively. In addition to the DC and AC motors mentioned here, control systems use brushless DC motors (Reliance Motion Control Corp., 1980) and stepping motors (Kuo, 1980). Models for these machines, developed in the works just cited, do not differ in principle from the motors considered in this section. In general, the analysis, supported by experiment, develops the torque as a function of voltage and speed similar to the AC motor torque–speed curves given in Fig. 2.33. From such curves one can obtain a linearized formula such as Eq. (2.61) to use in the mechanical part of the system and an equivalent circuit consisting of a resistance and an inductance to use in the electrical part.


2.4 Heat and Fluid-Flow Models Thermodynamics, heat transfer, and fluid dynamics are each the subject of complete textbooks. For purposes of generating dynamic models for use in control systems, the most important aspect of the physics is to represent the dynamic interaction between the variables. Experiments are usually required to determine the actual values of the parameters and thus to complete the dynamic model for purposes of control systems design.

2.4.1 Heat Flow Some control systems involve regulation of temperature for portions of the system. The dynamic models of temperature control systems involve the flow and storage of heat energy. Heat energy flows through substances at a rate proportional to the temperature difference across the substance; that is, q=

1 (T1 − T2 ), R

(2.62)

where q = heat energy flow, joules per second (J/sec), or British Thermal Unit/sec (BTU/sec), R = thermal resistance, ◦ C/J·sec or ◦ F/BTU·sec, T = temperature, ◦ C or ◦ F.

2.4 Heat and Fluid-Flow Models

51

The net heat-energy flow into a substance affects the temperature of the substance according to the relation 1 (2.63) T˙ = q, C where C is the thermal capacity. Typically, there are several paths for heat to flow into or out of a substance, and q in Eq. (2.63) is the sum of heat flows obeying Eq. (2.62).

EXAMPLE 2.14

Equations for Heat Flow A room with all but two sides insulated ( 1/R = 0) is shown in Fig. 2.34. Find the differential equations that determine the temperature in the room. Solution. Application of Eqs. (2.62) and (2.63) yields   1 1 1 (TO − TI ), T˙ I = + CI R1 R2 where CI = thermal capacity of air within the room, TO = temperature outside, TI = temperature inside, R2 = thermal resistance of the room ceiling, R1 = thermal resistance of the room wall. Normally the material properties are given in tables as follows:

Specific heat

1. The specific heat at constant volume cv , which is converted to heat capacity by C = mcv ,

Thermal conductivity

Figure 2.34 Dynamic model for room temperature

(2.64)

where m is the mass of the substance; 2. The thermal conductivity9 k, which is related to thermal resistance R by kA 1 = , R l where A is the cross-sectional area and l is the length of the heat-flow path.

q2 R2

Temperature outside, TO R1

q1

9 In the case of insulation for houses, resistance is quoted as R-values; for example, R-11 refers to a substance that has a resistance to heat flow equivalent to that given by 11 in. of solid wood.

52 Chapter 2 Dynamic Models In addition to flow due to transfer, as expressed by Eq. (2.62), heat can also flow when a warmer mass flows into a cooler mass, or vice versa. In this case, q = wcv (T1 − T2 ),

(2.65)

where w is the mass flow rate of the fluid at T1 flowing into the reservoir at T2 . For a more complete discussion of dynamic models for temperature control systems, see Cannon (1967) or textbooks on heat transfer.

EXAMPLE 2.15

Equations for Modeling a Heat Exchanger A heat exchanger is shown in Fig. 2.35. Steam enters the chamber through the controllable valve at the top, and cooler steam leaves at the bottom. There is a constant flow of water through the pipe that winds through the middle of the chamber so that it picks up heat from the steam. Find the differential equations that describe the dynamics of the measured water outflow temperature as a function of the area As of the steam-inlet control valve when open. The sensor that measures the water outflow temperature, being downstream from the exit temperature in the pipe, lags the temperature by td seconds. Solution. The temperature of the water in the pipe will vary continuously along the pipe as the heat flows from the steam to the water. The temperature of the steam will also reduce in the chamber as it passes over the maze of pipes. An accurate thermal model of this process is therefore quite involved because the actual heat transfer from the steam to the water will be proportional to the local temperatures of each fluid. For many control applications it is not necessary to have great accuracy because the feedback will correct for a considerable amount of error in the model. Therefore, it makes sense to combine the spatially varying temperatures into single temperatures Ts and Tw for the outflow steam and water temperatures, respectively. We then assume that the heat transfer from steam to water is proportional to the difference in these temperatures, as given by Eq. (2.62). There is also a flow of heat into the chamber

Figure 2.35 Heat exchanger

As Steam @ Tsi ws  KsAs

Water ww @ Twi Water @ Tw

Tm

Steam @ Ts

2.4 Heat and Fluid-Flow Models

53

from the inlet steam that depends on the steam flow rate and its temperature according to Eq. (2.65), qin = ws cvs (Tsi − Ts ), where ws = Ks As , mass flow rate of the steam, As = area of the steam inlet valve, Ks = flow coefficient of the inlet valve, cvs = specific heat of the steam, Tsi = temperature of the inflow steam, Ts = temperature of the outflow steam. The net heat flow into the chamber is the difference between the heat from the hot incoming steam and the heat flowing out to the water. This net flow determines the rate of temperature change of the steam according to Eq. (2.63), Cs T˙ s = As Ks cvs (Tsi − Ts ) −

1 (Ts − Tw ), R

(2.66)

where Cs = ms cvs is the thermal capacity of the steam in the chamber with mass ms , R = the thermal resistance of the heat flow averaged over the entire exchanger. Likewise, the differential equation describing the water temperature is Cw T˙ w = ww ccw (Twi − Tw ) +

1 (Ts − Tw ), R

(2.67)

where ww = mass flow rate of the water, ccw = specific heat of the water, Twi = temperature of the incoming water, Tw = temperature of the outflowing water. To complete the dynamics, the time delay between the measurement and the exit flow is described by the relation Tm (t) = Tw (t − td ), where Tm is the measured downstream temperature of the water and td is the time delay. There may also be a delay in the measurement of the steam temperature Ts , which would be modeled in the same manner. Equation (2.66) is nonlinear because the quantity Ts is multiplied by the control input As . The equation can be linearized about Tso (a specific value of Ts ) so that Tsi − Ts is assumed constant for purposes of approximating the nonlinear term, which we will define as Ts . In order to eliminate the Twi term in Eq. (2.67), it is convenient

54 Chapter 2 Dynamic Models to measure all temperatures in terms of deviation in degrees from Twi . The resulting equations are then 1 1 Cs T˙ s = − Ts + Tw + Ks cvs Ts As , R R   1 1 + ww cvw Tw + Ts , Cw T˙ w = − R R Tm = Tw (t − td ). Although the time delay is not a nonlinearity, we will see in Chapter 3 that operationally, Tm = e−td s Tw . Therefore, the transfer function of the heat exchanger has the form Ke−td s Tm (s) = . (2.68) As (s) (τ1 s + 1)(τ2 s + 1)

2.4.2 Incompressible Fluid Flow

The continuity relation

Fluid flows are common in many control systems components. One example is the hydraulic actuator, which is used extensively in control systems because it can supply a large force with low inertia and low weight. They are often used to move the aerodynamic control surfaces of airplanes, to gimbal rocket nozzles, to move the linkages in earth-moving equipment, farm tractor implements, snow-grooming machines, and to move robot arms. The physical relations governing fluid flow are continuity, force equilibrium, and resistance. The continuity relation is simply a statement of the conservation of matter; that is, (2.69) m ˙ = win − wout , where m = fluid mass within a prescribed portion of the system, win = mass flow rate into the prescribed portion of the system, wout = mass flow rate out of the prescribed portion of the system.

EXAMPLE 2.16

Equations for Describing Water Tank Height Determine the differential equation describing the height of the water in the tank in Fig. 2.36.

Figure 2.36 Water tank example

in

h

Pressure p1 out

2.4 Heat and Fluid-Flow Models

55

Solution. Application of Eq. (2.69) yields 1 h˙ = (win − wout ) , Aρ

(2.70)

where A = area of the tank, ρ = density of water, h = m/Aρ = height of water, m = mass of water in the tank. Force equilibrium must apply exactly as described by Eq. (2.1) for mechanical systems. Sometimes in fluid-flow systems some forces result from fluid pressure acting on a piston. In this case the force from the fluid is f = pA,

(2.71)

where f = force, p = pressure in the fluid, A = area on which the fluid acts.

EXAMPLE 2.17

Modeling a Hydraulic Piston Determine the differential equation describing the motion of the piston actuator shown in Fig. 2.37, given that there is a force FD acting on it and a pressure p in the chamber. Solution. Equations (2.1) and (2.71) apply directly, where the forces include the fluid pressure as well as the applied force. The result is M x¨ = Ap − FD , where A = area of the piston, p = pressure in the chamber,

Figure 2.37 Hydraulic piston actuator

x

Piston

FD

Liquid at pressure p

56 Chapter 2 Dynamic Models M = mass of the piston, x = position of the piston. In many cases of fluid-flow problems the flow is resisted either by a constriction in the path or by friction. The general form of the effect of resistance is given by 1 (2.72) w = (p1 − p2 )1/α , R where w = mass flow rate, p1 , p2 = pressures at ends of the path through which flow is occurring, R, α = constants whose values depend on the type of restriction. Or, as is more commonly used in hydraulics, 1 (p1 − p2 )1/α , Q= ρR where

(2.73)

Q = volume flow rate, where Q = w/ρ , ρ = fluid density. The constant α takes on values between 1 and 2. The most common value is approximately 2 for high flow rates (those having a Reynolds number Re > 105 ) through pipes or through short constrictions or nozzles. For very slow flows through long pipes or porous plugs wherein the flow remains laminar (Re
1000), α = 1. Flow rates between these extremes can yield intermediate values of α. The Reynolds number indicates the relative importance of inertial forces and viscous forces in the flow. It is proportional to a material’s velocity and density and to the size of the restriction, and it is inversely proportional to the viscosity. When Re is small, the viscous forces predominate and the flow is laminar. When Re is large, the inertial forces predominate and the flow is turbulent. Note that a value of α = 2 indicates that the flow is proportional to the square root of the pressure difference and therefore will produce a nonlinear differential equation. For the initial stages of control systems analysis and design, it is typically very useful to linearize these equations so that the design techniques described in this book can be applied. Linearization involves selecting an operating point and expanding the nonlinear term to be a small perturbation from that point.

EXAMPLE 2.18

Linearization of Water Tank Height and Outflow Find the nonlinear differential equation describing the height of the water in the tank in Fig. 2.36. Assume that there is a relatively short restriction at the outlet and that α = 2. Also linearize your equation about the operating point ho . Solution. Applying Eq. (2.72) yields the flow out of the tank as a function of the height of the water in the tank: 1 wout = (p1 − pa )1/2 . (2.74) R

2.4 Heat and Fluid-Flow Models

57

Here, p1 = ρgh + pa , the hydrostatic pressure, pa = ambient pressure outside the restriction. Substituting Eq. (2.74) into Eq. (2.70) yields the nonlinear differential equation for the height:   √ ˙h = 1 win − 1 p1 − pa . (2.75) Aρ R Linearization involves selecting the operating point po = ρgho + pa and substituting p1 = po + p into Eq. (2.74). Then we expand the nonlinear term according to the relation (1 + ε)β ∼ (2.76) = 1 + βε, where ε 1. Equation (2.74) can thus be written as 1/2  √ p po − pa 1+ wout = R po − pa   √ 1 p po − pa ∼ . 1+ = R 2 po − pa

(2.77)

The linearizing approximation made in Eq. (2.77) is valid as long as p po − pa ; that is, as long as the deviations of the system pressure from the chosen operating point are relatively small. Combining Eqs. (2.70) and (2.77) yields the following linearized equation of motion for the water tank level:    √ po − pa 1 1 p ˙ h = . win − 1+ Aρ R 2 po − pa Because p = ρg h, this equation reduces to h˙ = −

win g − h + √ 2AR po − pa Aρ

√

po − pa , ρAR

(2.78)

˙ The operating point is not an equilibrium which is a linear differential equation for h. point because some control input is required to maintain it. In other words, when the system is at the operating point ( h = 0) with no input (win = 0), it will move from that point because h˙ = 0. So if no water is flowing into the tank, the tank will drain, thus moving it from the reference point. To define an operating point that is also an equilibrium point, we need to require that there be a nominal flow rate, √ wino po − pa = , Aρ ρAR and define the linearized input flow to be a perturbation from that value. Hydraulic actuators

Hydraulic actuators obey the same fundamental relationships we saw in the water tank: continuity [Eq. (2.69)], force balance [Eq. (2.71)], and flow resistance [Eq. (2.72)]. Although the development here assumes the fluid is perfectly incompressible, in fact, hydraulic fluid has some compressibility due primarily to entrained

58 Chapter 2 Dynamic Models air. This feature causes hydraulic actuators to have some resonance because the compressibility of the fluid acts like a stiff spring. This resonance limits their speed of response.

EXAMPLE 2.19

Modeling a Hydraulic Actuator 1. Find the nonlinear differential equations relating the movement θ of the control surface to the input displacement x of the valve for the hydraulic actuator shown in Fig. 2.38. 2. Find the linear approximation to the equations of motion when y˙ = constant, with and without an applied load—that is, when F = 0 and when F = 0. Assume that θ motion is small. Solution 1. Equations of motion: When the valve is at x = 0, both passages are closed and no motion results. When x > 0, as shown in Fig. 2.38, the oil flows clockwise as shown and the piston is forced to the left. When x < 0, the fluid flows counterclockwise. The oil supply at high pressure ps enters the left side of the large piston chamber, forcing the piston to the right. This causes the oil to flow out of the valve chamber from the rightmost channel instead of the left. We assume that the flow through the orifice formed by the valve is proportional to x; that is, 1 (ps − p1 )1/2 x. (2.79) Q1 = ρR1 Similarly, 1 Q2 = (p2 − pe )1/2 x. (2.80) ρR2 1

Valve

2

x

pe

ps

2

pe 1

u p1 F

p2

Fa y l Low-pressure oil

Figure 2.38 Hydraulic actuator with valve

High-pressure oil

d

Aerodynamic control surface

2.4 Heat and Fluid-Flow Models

59

The continuity relation yields A˙y = Q1 = Q2 ,

(2.81)

where A = piston area. The force balance on the piston yields A(p1 − p2 ) − F = m¨y,

(2.82)

where m = mass of the piston and the attached rod, F = force applied by the piston rod to the control surface attachment point. Furthermore, the moment balance of the control surface using Eq. (2.14) yields I θ¨ = Fl cos θ − Fa d,

(2.83)

where I = moment of inertia of the control surface and attachment about the hinge, Fa = applied aerodynamic load. To solve this set of five equations, we require the following additional kinematic relationship between θ and y: y = l sin θ.

(2.84)

The actuator is usually constructed so that the valve exposes the two passages equally; therefore, R1 = R2 , and we can infer from Eqs. (2.79) to (2.81) that ps − p1 = p2 − pe .

(2.85)

These relations complete the nonlinear differential equations of motion; they are formidable and difficult to solve. 2. Linearization and simplification: For the case in which y˙ = a constant (¨y = 0) and there is no applied load (F = 0), Eqs. (2.82) and (2.85) indicate that p s + pe . (2.86) 2 Therefore, using Eq. (2.81) and letting sin θ = θ (since θ is assumed to be small), we get √ ps − p e θ˙ = √ x. (2.87) 2AρRl This represents a single integration between the input x and the output θ, where the proportionality constant is a function only of the supply pressure and the fixed parameters of the actuator. For the case y˙ = constant but F = 0, Eqs. (2.82) and (2.85) indicate that ps + pe + F/A p1 = 2 p1 = p2 =

60 Chapter 2 Dynamic Models and

√

ps − pe − F/A x. (2.88) √ 2Aρ Rl This result is also a single integration between the input x and the output θ, but the proportionality constant now depends on the applied load F. As long as the commanded values of x produce θ motion that has a suf¨ the approximation given by Eqs. (2.87) or (2.88) is ficiently small value of θ, valid and no other linearized dynamic relationships are necessary. However, as soon as the commanded values of x produce accelerations in which the iner¨ are a significant fraction of ps − pe , the tial forces (m¨y and the reaction to I θ) approximations are no longer valid. We must then incorporate these forces into the equations, thus obtaining a dynamic relationship between x and θ that is much more involved than the pure integration implied by Eqs. (2.87) or (2.88). Typically, for initial control system designs, hydraulic actuators are assumed to obey the simple relationship of Eqs. (2.87) or (2.88). When hydraulic actuators are used in feedback control systems, resonances have been encountered that are not explained by using the approximation that the device is a simple integrator as in Eqs. (2.87) or (2.88). The source of the resonance is the neglected accelerations discussed above along with the additional feature that the oil is slightly compressible due to small quantities of entrained air. This phenomenon is called the “oil-mass resonance.” θ˙ =

2.5 Historical Perspective Newton’s second law of motion (Eq. 2.1) was first published in his Philosophiae Naturalis Principia Mathematica in 1686 along with his two other famous laws of motion. The first: A body will continue with the same uniform motion unless acted on by an external unbalanced force, and the third: To every action there is an equal and opposite reaction. Isaac Newton also published his law of gravitation in this same publication, which stated that every mass particle attracts all other particles by a force proportional to the inverse of the square of the distance between them and the product of their two masses. His basis for developing these laws was the work of several other early scientists, combined with his own development of the calculus in order to reconcile all the observations. It is amazing that these laws still stand today as the basis for almost all dynamic analysis with the exception of Einstein’s additions in the early 1900s for relativistic effects. It is also amazing that Newton’s development of calculus formed the foundation of our mathematics that enable dynamic modeling. In addition to being brilliant, he was also very eccentric. As Brennan writes in Heisenberg Probably Slept Here, “He was seen about campus in his disheveled clothes, his wig askew, wearing run-down shoes and a soiled neckpiece. He seemed to care about nothing but his work. He was so absorbed in his studies that he forgot to eat.” Another interesting aspect of Newton is that he initially developed the calculus and the now famous laws of physics about 20 years prior to publishing them! The incentive to publish them arose from a bet between three men having lunch at a pub in 1684: Edmond Halley, Christopher Wren, and Robert Hooke. They all had the opinion that

2.5 Historical Perspective

61

Kepler’s elliptical characterization of planetary motion could be explained by the inverse square law, but nobody had ever proved it, so they “placed a bet as to who could first prove the conjecture.”10 Halley went to Newton for help due to his fame as a mathematician, who responded he had already done it many years ago and would forward the papers to him. He not only did that shortly afterwards, but followed it up with the Principia with all the details two years later. The basis for Newton’s work started with the astronomer Nicholas Copernicus more than a hundred years before the Principia was published. He was the first to speculate that the planets revolved around the sun, rather than everything in the skies revolving around the earth. But Copernicus’heretical notion was largely ignored at the time, except by the church who banned his publication. However, two scientists did take note of his work: Galileo Galilei in Italy and Johannes Kepler in Austria. Kepler relied on a large collection of astronomical data taken by a Danish astronomer, Tycho Brahe, and concluded that the planetary orbits were ellipses rather than the circles that Copernicus had postulated. Galileo was an expert telescope builder and was able to clearly establish that the earth was not the center of all motion, partly because he was able to see moons revolving around other planets. He also did experiments with rolling balls down inclined planes that strongly suggested that F = ma (alas, it’s a myth that he did his experiments by dropping objects out of the Leaning Tower of Pisa). Galileo published his work in 1632, which raised the ire of the church who then later banned him to house arrest until he died.11 It was not until 1985 that the church recognized the important contributions of Galileo! These men laid the groundwork for Newton to put it all together with his laws of motion and the inverse square gravitational law. With these two physical principles, all the observations fit together with a theoretical framework that today forms the basis for the modeling of dynamic systems. The sequence of discoveries that ultimately led to the laws of dynamics that we take for granted today were especially remarkable when we stop to think that they were all carried out without a computer, a calculator, or even a slide rule. On top of that, Newton had to invent calculus in order to reconcile the data. After publishing the Principia, Newton went on to be elected to Parliament and was given high honors, including being the first man of science to be knighted by the Queen. He also got into fights with other scientists fairly regularly and used his powerful positions to get what he wanted. In one instance, he wanted data from the Royal Observatory that was not forthcoming fast enough. So he created a new board with authority over the Observatory and had the Astronomer Royal expelled from the Royal Society. Newton also had other less scientific interests. Many years after his death, John Maynard Keynes found that Newton had been spending as much of his time on metaphysical occult, alchemy, and biblical works as he had been on physics. More than a hundred years after Newton’s Principia, Michael Faraday performed a multitude of experiments and postulated the notion of electromagnetic lines of force in free space. He also discovered induction (Faraday’s Law), which led to the 10 Much of the background on Newton was taken from Heisenberg Probably Slept Here, by Richard

P. Brennan, 1997. The book discusses his work and the other early scientists that laid the groundwork for Newton. 11 Galileo’s life, accomplishments, and house arrest are very well described in Dava Sobel’s book, Galileo’s Daughter.

62 Chapter 2 Dynamic Models electric motor and the laws of electrolysis. Faraday was born into a poor family, had virtually no schooling, and became an apprentice to a bookbinder at age 14. There he read many of the books being bound and became fascinated by science articles. Enthralled by these, he maneuvered to get a job as a bottle washer for a famous scientist, eventually learned enough to be a competitor to him, and ultimately became a professor at the Royal Institution in London. But lacking a formal education, he had no mathematical skills, and lacked the ability to create a theoretical framework for his discoveries. Faraday became a famous scientist in spite of his humble origins. After he had achieved fame for his discoveries and was made a Fellow of the Royal Society, the prime minister asked him what good his inventions could be.12 Faraday’s answer was, “Why Prime Minister, someday you can tax it.” But in those days, scientists were almost exclusively men born into privilege; so Faraday had been treated like a second-class citizen by some of the other scientists. As a result, he rejected knighthood as well as burial at Westminster Abbey. Faraday’s observations, along with those by Coulomb and Ampere, led James Clerk Maxwell to integrate all their knowledge on magnetism and electricity into Maxwell’s Equations. Against the beliefs of most prominent scientists of the day (Faraday being an exception), Maxwell invented the concepts of fields and waves that explained magnetic and electrostatic forces and was the key to creating the unifying theory. Although Newton had discovered the spectrum of light, Maxwell was also the first to realize that light was one type of the same electromagnetic waves, and its behavior was explained as well by Maxwell’s Equations. In fact, the only constant in his equations are µ and ε. The constant speed √ of light is c = 1/ µε. Maxwell was a Scottish mathematician and theoretical physicist. His work has been called the second great unification in physics, the first being that due to Newton. Maxwell was born into the privileged class and was given the benefits of an excellent education and he excelled at it. In fact, he was an extremely gifted theoretical and experimental scientist as well as a very generous and kind man with many friends and little vanity. In addition to unifying the observations of electromagnetics into a theory that still governs our engineering analyses today, he was the first to present an explanation of how light travels, the primary colors, the kinetic theory of gases, the stability of Saturn’s rings, and the stability of feedback control systems! His discovery of the three primary colors (red, green, and blue) forms the basis of our color television to this day. His theory showing the speed of light is a constant was difficult to reconcile with Newton’s laws and led Albert Einstein to create the special theory of relativity in the early 1900s. This led Einstein to say, “One scientific epoch ended and another began with James Clerk Maxwell.”13

SUMMARY Mathematical modeling of the system to be controlled is the first step in analyzing and designing the required system controls. In this chapter we developed models 12 E = MC 2 , A Biography of the World’s Most Famous Equation, by David Bodanis, Walker and Co., New York, 2000. 13 The Man Who Changed Everything: The Life of James Clerk Maxwell, Basil Mahon, Wiley, 2003.

Review Questions

TABLE 2.1

63

Key Equations for Dynamic Models System Mechanical

Electrical Electromechanical

Back emf Heat flow

Fluid flow

Important Laws or Relationships

Associated Equations

Translation motion (Newton’s law) Rotational motion Operational amplifier Law of motors Law of the generator Torque developed in a rotor Voltage generated as a result of rotation of a rotor Heat-energy flow Temperature as a function of heat-energy flow Specific heat

F = ma

(2.1)

M = Iα

(2.14) (2.36), (2.37) (2.43) (2.46) (2.50) (2.51)

Continuity relation (conservation of matter) Force of a fluid acting on a piston Effect of resistance to fluid flow

F = Bli e(t) = Blv T = Kt ia e = Ke θ˙m

Equation Number

q = 1/R(T1 − T2 ) T˙ = 1/C q

(2.62) (2.63)

C = mcv

(2.64)

˙ = win − wout m

(2.69)

f = pA

(2.71)

w = 1/R(p1 − p2 )1/α

(2.72)

for representative systems. Important equations for each category of system are summarized in Table 2.1.

REVIEW QUESTIONS 1. What is a “free-body diagram”? 2. What are the two forms of Newton’s law? 3.

For a structural process to be controlled, such as a robot arm, what is the meaning of “collocated control”? “Noncollocated control”?

4.

State Kirchhoff’s current law.

5.

State Kirchhoff’s voltage law.

6. When, why, and by whom was the device named an “operational amplifier”? 7. What is the major benefit of having zero input current to an operational amplifier? 8. Why is it important to have a small value for the armature resistance Ra of an electric motor? 9. What are the definition and units of the electric constant of a motor? 10. What are the definition and units of the torque constant of an electric motor?

64 Chapter 2 Dynamic Models 11. Why do we approximate a physical model of the plant (which is always nonlinear) with a linear model?



12. Give the relationships for (a) heat flow across a substance, and (b) heat storage in a substance.



13. Name and give the equations for the three relationships governing fluid flow.

PROBLEMS Problems for Section 2.1: Dynamics of Mechanical Systems 2.1 Write the differential equations for the mechanical systems shown in Fig. 2.39. For (a) and (b), state whether you think the system will eventually decay so that it has no motion at all, given that there are non-zero initial conditions for both masses, and give a reason for your answer.

Figure 2.39 Mechanical systems

x1

b1 m1

x2 k2

m2

y

k3

k1 No friction

Friction, b2

(a) k1

m1

k2

Friction, b1

m2

k3

No friction (b) b1

k1

m1

m2

F

k2 No friction

No friction

(c)

2.2 Write the differential equation for the mechanical system shown in Fig. 2.40. State whether you think the system will eventually decay so that it has no motion at all, given that there are non-zero initial conditions for both masses, and give a reason for your answer. 2.3 Write the equations of motion for the double-pendulum system shown in Fig. 2.41. Assume that the displacement angles of the pendulums are small enough to ensure that the spring is always horizontal. The pendulum rods are taken to be massless, of length l, and the springs are attached three-fourths of the way down.

Problems Figure 2.40 Mechanical system for Problem 2.2

x1 K1

65

x2 b2

m1

K1 m1

K2 No friction

No friction

Figure 2.41 Double pendulum k

m

m

2.4 Write the equations of motion of a pendulum consisting of a thin, 4 kg stick of length l suspended from a pivot. How long should the rod be in order for the period to be exactly 2 sec? (The inertia I of a thin stick about an end point is 13 ml2 . Assume that θ is small enough that sin θ ∼ = θ.) Why do you think grandfather clocks are typically about 6 ft high? 2.5

For the car suspension discussed in Example 2.2, plot the position of the car and the wheel after the car hits a “unit bump”(i.e., r is a unit step) using MATLAB. Assume that m1 = 10 kg, m2 = 350 kg, Kw = 500,000 N/m, Ks = 10,000 N/m. Find the value of b that you would prefer if you were a passenger in the car.

2.6 Write the equations of motion for a body of mass M suspended from a fixed point by a spring with a constant k. Carefully define where the body’s displacement is zero. 2.7 Automobile manufacturers are contemplating building active suspension systems. The simplest change is to make shock absorbers with a changeable damping, b(u1 ). It is also possible to make a device to be placed in parallel with the springs that has the ability to supply an equal force, u2, in opposite directions on the wheel axle and the car body. (a) Modify the equations of motion in Example 2.2 to include such control inputs. (b) Is the resulting system linear? (c) Is it possible to use the forcer u2 to completely replace the springs and shock absorber? Is this a good idea? 2.8 Modify the equation of motion for the cruise control in Example 2.1, Eq. (2.4), so that it has a control law; that is, let u = K(vr − v), (2.89) where vr = reference speed,

(2.90)

K = constant.

(2.91)

This is a “proportional”control law in which the difference between vr and the actual speed is used as a signal to speed the engine up or slow it down. Revise the equations of motion with vr as the input and v as the output and find the transfer function. Assume that m = 1000 kg and b = 50 N·sec/m, and find the response for a unit step in vr using

66 Chapter 2 Dynamic Models MATLAB. Using trial and error, find a value of K that you think would result in a control system in which the actual speed converges as quickly as possible to the reference speed with no objectionable behavior. 2.9 In many mechanical positioning systems there is flexibility between one part of the system and another. An example is shown in Fig. 2.7 where there is flexibility of the solar panels. Fig. 2.42 depicts such a situation, where a force u is applied to the mass M and another mass m is connected to it. The coupling between the objects is often modeled by a spring constant k with a damping coefficient b, although the actual situation is usually much more complicated than this. (a) Write the equations of motion governing this system. (b) Find the transfer function between the control input u and the output y.

Figure 2.42 Schematic of a system with flexibility

y

x k m

M

u

b

Problems for Section 2.2: Models of Electric Circuits 2.10 A first step toward a realistic model of an op-amp is given by the following equations and is shown in Fig. 2.43: 107 [V+ − V− ], s+1 i+ = i− = 0.

Vout =

Find the transfer function of the simple amplification circuit shown using this model.

Figure 2.43 Circuit for Problem 2.10

Rf Rin

Vin

y
y




Vout

2.11 Show that the op-amp connection shown in Fig. 2.44 results in Vout = Vin if the op-amp is ideal. Give the transfer function if the op-amp has the nonideal transfer function of Problem 2.10. 2.12 Show that, with the nonideal transfer function of Problem 2.10, the op-amp connection shown in Fig. 2.45 is unstable.

Problems Figure 2.44 Circuit for Problem 2.11

y
y

Vin

Figure 2.45 Circuit for Problem 2.12

y y


Vin

67




Vout




Vout

2.13 A common connection for a motor power amplifier is shown in Fig. 2.46. The idea is to have the motor current follow the input voltage, and the connection is called a current amplifier. Assume that the sense resistor rs is very small compared with the feedback resistor R, and find the transfer function from Vin to Ia . Also show the transfer function when Rf = ∞.

Figure 2.46 Op-amp circuit for Problem 2.13

Rf Rin

y
R

y




Vin

Ia




DC motor Vout Rs

2.14 An op-amp connection with feedback to both the negative and the positive terminals is shown in Fig. 2.47. If the op-amp has the nonideal transfer function given in Problem 2.10, r , in terms of give the maximum value possible for the positive feedback ratio, P = r+R

in , for the circuit to remain stable. the negative feedback ratio, N = RinR+R f 2.15 Write the dynamic equations and find the transfer functions for the circuits shown in Fig. 2.48.

(a) (b) (c) (d)

passive lead circuit active lead circuit active lag circuit passive notch circuit

68 Chapter 2 Dynamic Models Figure 2.47 Op-amp circuit for Problem 2.14

Rf Rin

y
y




Vin

R

Vout

r

Figure 2.48 (a) Passive lead; (b) active lead; (c) active lag; and (d) passive notch circuits

C

R1

u

R2

y

(a) Rf

R2

R1

Vin

Vout

C (b) R2

R1

C

Rin Vin

Vout (c) C

C

R

R Vin

2C

R/2

Vout

(d)

2.16 The very flexible circuit shown in Fig. 2.49 is called a biquad because its transfer function can be made to be the ratio of two second-order or quadratic polynomials. By selecting different values for Ra , Rb , Rc , and Rd , the circuit can realize a low-pass, band-pass, high-pass, or band-reject (notch) filter.

Problems

69

C C R R1 Vin

R

R2 R


V1

R







V2

V3

R

Ra Rb

Vout

Rc Rd

Figure 2.49 Op-amp biquad (a) Show that if Ra = R, and Rb = Rc = Rd = ∞, the transfer function from Vin to Vout can be written as the low-pass filter Vout = 2 Vin s

A

s + 2ζ +1 2 ωn ωn

,

(2.92)

where A=

R , R1

1 , RC R . ζ = 2R2

ωn =

(b) Using the MATLAB command step, compute and plot on the same graph the step responses for the biquad of Fig. 2.49 for A = 1, ωn = 1, and ζ = 0.1, 0.5, and 1.0. 2.17 Find the equations and transfer function for the biquad circuit of Fig. 2.49 if Ra = R, Rd = R1 , and Rb = Rc = ∞.

Problems for Section 2.3: Models of Electromechanical Systems 2.18 The torque constant of a motor is the ratio of torque to current and is often given in ounceinches per ampere. (Ounce-inches have dimension force × distance, where an ounce is 1/16 of a pound.) The electric constant of a motor is the ratio of back emf to speed and is often given in volts per 1000 rpm. In consistent units, the two constants are the same for a given motor. (a) Show that the units ounce-inches per ampere are proportional to volts per 1000 rpm by reducing both to MKS (SI) units.

70 Chapter 2 Dynamic Models (b) A certain motor has a back emf of 25 V at 1000 rpm. What is its torque constant in ounce-inches per ampere? (c) What is the torque constant of the motor of part (b) in newton-meters per ampere? 2.19 The electromechanical system shown in Fig. 2.50 represents a simplified model of a capacitor microphone. The system consists in part of a parallel plate capacitor connected into an electric circuit. Capacitor plate a is rigidly fastened to the microphone frame. Sound waves pass through the mouthpiece and exert a force fs (t) on plate b, which has mass M and is connected to the frame by a set of springs and dampers. The capacitance C is a function of the distance x between the plates, as follows: C(x) =

εA , x

where ε = dielectric constant of the material between the plates, A = surface area of the plates. The charge q and the voltage e across the plates are related by q = C(x)e. The electric field in turn produces the following force fe on the movable plate that opposes its motion: q2 fe = . 2εA (a) Write differential equations that describe the operation of this system. (It is acceptable to leave in nonlinear form.) (b) Can one get a linear model? (c) What is the output of the system?

Figure 2.50 Simplified model for capacitor microphone

L

R i(t)

b K B

A M

a

 y


e

fs(t) x

2.20 A very typical problem of electromechanical position control is an electric motor driving a load that has one dominant vibration mode. The problem arises in computer-disk-head control, reel-to-reel tape drives, and many other applications. A schematic diagram is sketched in Fig. 2.51. The motor has an electrical constant Ke , a torque constant Kt , an armature inductance La , and a resistance Ra . The rotor has an inertia J1 and a viscous friction B. The load has an inertia J2 . The two inertias are connected by a shaft with a spring constant k and an equivalent viscous damping b. Write the equations of motion.

Problems Figure 2.51 Motor with a flexible load

La

Ra u1

 ya

u2

k, b




J2

J1



71

Problems for Section 2.4: Heat and Fluid-Flow Models 2.21 A precision table-leveling scheme shown in Fig. 2.52 relies on thermal expansion of actuators under two corners to level the table by raising or lowering their respective corners. The parameters are as follows: Tact = actuator temperature, Tamb = ambient air temperature, Rf = heat-flow coefficient between the actuator and the air, C = thermal capacity of the actuator, R = resistance of the heater. Assume that (1) the actuator acts as a pure electric resistance, (2) the heat flow into the actuator is proportional to the electric power input, and (3) the motion d is proportional to the difference between Tact and Tamb due to thermal expansion. Find the differential equations relating the height of the actuator d versus the applied voltage vi .

Figure 2.52 (a) Precision table kept level by actuators; (b) side view of one actuator

d

yi

Tact yi

(a)

(b)

2.22 An air conditioner supplies cold air at the same temperature to each room on the fourth floor of the high-rise building shown in Fig. 2.53(a). The floor plan is shown in Fig. 2.53(b). The cold airflow produces an equal amount of heat flow q out of each room. Write a set of differential equations governing the temperature in each room, where To = temperature outside the building, Ro = resistance to heat flow through the outer walls, Ri = resistance to heat flow through the inner walls.

72 Chapter 2 Dynamic Models Figure 2.53 Building air conditioning: (a) high-rise building; (b) floor plan of the fourth floor

Ro Ri

Fourth floor

(b)

(a)

Assume that (1) all rooms are perfect squares, (2) there is no heat flow through the floors or ceilings, and (3) the temperature in each room is uniform throughout the room. Take advantage of symmetry to reduce the number of differential equations to three. 2.23 For the two-tank fluid-flow system shown in Fig. 2.54, find the differential equations relating the flow into the first tank to the flow out of the second tank.

Figure 2.54 Two-tank fluid-flow system for Problem 2.23

win

wout

2.24 A laboratory experiment in the flow of water through two tanks is sketched in Fig. 2.55. Assume that Eq. (2.74) describes flow through the equal-sized holes at points A, B, or C. (a) With holes at A and C, but none at B, write the equations of motion for this system in terms of h1 and h2 . Assume that h3 = 20 cm, h1 > 20 cm, and h2 < 20 cm. When h2 = 10 cm, the outflow is 200 g/min. (b) At h1 = 30 cm and h2 = 10 cm, compute a linearized model and the transfer function from pump flow (in cubic centimeters per minute) to h2 . (c) Repeat parts (a) and (b) assuming hole A is closed and hole B is open.

Problems Figure 2.55 Two-tank fluid-flow system for Problem 2.24

73

Pump

A

h1

h3 h2 B

C

2.25 The equations for heating a house are given by Eqs. (2.62) and (2.63), and in a particular case can be written with time in hours as C

T − To dTh = Ku − h , dt R

where (a) (b) (c) (d) (e) (f)

C is the thermal capacity of the house, BTU/◦ F, Th is the temperature in the house, ◦ F, To is the temperature outside the house, ◦ F, K is the heat rating of the furnace, = 90,000 BTU/hour, R is the thermal resistance, ◦ F per BTU/hour, u is the furnace switch, = 1 if the furnace is on and = 0 if the furnace is off.

It is measured that, with the outside temperature at 32◦ F and the house at 60◦ F, the furnace raises the temperature 2◦ F in 6 minutes (0.1 hour). With the furnace off, the house temperature falls 2◦ F in 40 minutes. What are the values of C and R for the house?

3

Dynamic Response 2.0

z0 0.1 0.2 0.3 0.4 0.5 0.6

1.8 1.6 1.4 1.2 y(t)

1.0 0.8 z  0.7 0.8 0.9 1.0

0.6 0.4 0.2 0

0

2

4

6 vnt

8

10

12

A Perspective on System Response We saw in Chapter 2 how to obtain the dynamic model of a system. In designing a control system, it is important to see how well a trial design matches the desired performance. We do this by solving the equations of the system model. There are two ways to approach solving the dynamic equations. For a quick, approximate analysis we use linear analysis techniques. The resulting approximations of system response provide insight into why the solution has certain features and how the system might be changed to modify the response in a desired direction. In contrast, a precise picture of the system response typically calls for numerical simulation of nonlinear equations of motion using computer aids. This chapter focuses on linear analysis and computer tools that can be used to solve for the time response of linear systems. There are three domains within which to study dynamic response: the s-plane, the frequency response, and the state space (analysis using the state-variable description). The well-prepared control engineer needs to be fluent in all of them, so they will be treated in depth in Chapters 5, 6, and 7, respectively. The purpose of this chapter is to discuss some of the fundamental

74

3.1 Review of Laplace Transforms

75

mathematical tools needed before studying analysis in the s-plane, frequency response, and state space.

Chapter Overview The Laplace transform, reviewed in Section 3.1 (and Appendix A), is the mathematical tool for transforming differential equations into an easier-tomanipulate algebraic form. In addition to the mathematical tools at our disposal, there are graphical tools that can help us to visualize the model of a system and evaluate the pertinent mathematical relationships between elements of the system. One approach is the block diagram, which was introduced in Chapter 1. Block diagram manipulation is discussed in Section 3.2 and allows the determination of transfer functions. Once the transfer function has been determined, we can identify its poles and zeros, which tell us a great deal about system characteristics, including its frequency response introduced in Section 3.1. Sections 3.3 to 3.5 focus on poles and zeros and some of the ways for manipulating them to steer system characteristics in a desired way. When feedback is introduced, the possibility that the system may become unstable is introduced. To study this effect, in Section 3.6 we consider the definition of stability and Routh’s test, which can determine stability by examining the coefficients of the system’s characteristic equation. Development of a model based on experimental time– response data is discussed in Section 3.7. Section 3.8 discusses amplitude and time scaling. Finally, Section 3.9 provides the historical perspective for the material in this chapter. An alternative representation of a system in graphical form is the signal-flow graph and flow graphs that allow the determination of complicated transfer functions are discussed in Appendix W3 on the web.

3.1 Review of Laplace Transforms Two attributes of linear time-invariant systems (LTIs) form the basis for almost all analytical techniques applied to these systems: 1. A linear system response obeys the principle of superposition. 2. The response of an LTI system can be expressed as the convolution of the input with the unit impulse response of the system. The concepts of superposition, convolution, and impulse response will be defined shortly. From the second property (as we will show), it follows immediately that the response of an LTI system to an exponential input is also exponential. This result is the principal reason for the usefulness of Fourier and Laplace transforms in the study of LTI systems.

3.1.1 Response by Convolution Superposition

The principle of superposition states that if the system has an input that can be expressed as a sum of signals, then the response of the system can be expressed

76 Chapter 3 Dynamic Response as the sum of the individual responses to the respective signals. We can express superposition mathematically. Consider the system to have input u and output y. Suppose further that, with the system at rest, we apply the input u1 (t) and observe the output y1 (t). After restoring the system to rest, we apply a second input u2 (t) and again observe the output, which we call y2 (t). Then, we form the composite input u(t) = α1 u1 (t) + α2 u2 (t). Finally, if superposition applies, then the response will be y(t) = α1 y1 (t) + α2 y2 (t). Superposition will apply if and only if the system is linear.

EXAMPLE 3.1

Superposition Show that superposition holds for the system modeled by the first-order linear differential equation y˙ + ky = u. Solution. We let u = α1 u1 + α2 u2 and assume that y = α1 y1 + α2 y2 . Then y˙ = α1 y˙ 1 + α2 y˙ 2 . If we substitute these expressions into the system equation, we get α1 y˙ 1 + α2 y˙ 2 + k(α1 y1 + α2 y2 ) = α1 u1 + α2 u2 . From this it follows that α1 (˙y1 + ky1 − u1 ) + α2 (˙y2 + ky2 − u2 ) = 0.

(3.1)

If y1 is the solution with input u1 and y2 is the solution with input u2 , then Eq. (3.1) is satisfied, the response is the sum of the individual responses, and superposition holds. Notice that the superposition result of Eq. (3.1) would also hold if k were a function of time. If it were constant, we call the system time invariant. In that case, it follows that if the input is delayed or shifted in time, then the output is unchanged except also being shifted by exactly the same amount. Mathematically, this is expressed by saying that, if y1 (t) is the output caused by u1 (t) then y1 (t − τ ) will be the response to u1 (t − τ ).

EXAMPLE 3.2

Time Invariance Consider y˙ 1 (t) + k(t)y1 (t) = u1 (t) and y˙ 2 (t) + k(t)y2 (t) = u1 (t − τ ), where τ is a constant shift. Assume that y2 (t) = y1 (t − τ ); then dy1 (t − τ ) + k(t)y1 (t − τ ) = u1 (t − τ ). dt Let us make the change of variable t − τ = η, then dy1 (η) + k(η + τ )y1 (η) = u1 (η). dη

(3.2)

3.1 Review of Laplace Transforms

77

If k(η + τ ) = k = constant, then dy1 (η) + ky1 (η) = u(η), dη which is Eq. (3.1). Therefore, we conclude that if the system is time invariant y(t − τ ) will be the response to u(t − τ ); that is if the input is delayed by τ sec, then the output is also delayed by τ sec.

Short pulse

We are able to solve for the response of a linear system to a general signal simply by decomposing the given signal into a sum of the elementary components and, by superposition, concluding that the response to the general signal is the sum of the responses to the elementary signals. In order for this process to work, the elementary signals need to be sufficiently “rich” that any reasonable signal can be expressed as a sum of them, and their responses have to be easy to find. The most common candidates for elementary signals for use in linear systems are the impulse and the exponential. Suppose the input signal to an LTI system is u1 (t) = p(t), and the corresponding output signal is y1 (t) = h(t) as shown in Fig. 3.1(a). Now if the input is scaled to u1 (t) = u(0)p(t), then by the scaling property of superposition, the output response will be y1 (t) = u(0)h(t). We showed that an LTI system obeys time invariance. If we delay the short pulse signal in time by τ , then the input is of the form u2 (t) = p(t − τ ) and the output response will also be delayed by the same amount y2 (t) = h(t − τ ) as shown in Fig. 3.1(b). Now by superposition, the response to the two short pulses will be the sum of the two individual outputs as shown in Fig. 3.1(c). If we have four pulses as the input, then the output will be the sum of the four individual responses as shown in Fig. 3.1(d). Any arbitrary input signal u(t) may be approximated by a series of pulses as shown in Fig. 3.2. We define a short pulse p (t) as a rectangular pulse having unit area such that

1 , 0 ≤ t ≤ p (t) = (3.3) 0, elsewhere as shown in Fig. 3.1(a). Suppose the response of the system to p (t) is defined as h (t). The response at time n to u(k )p (k ) is u(k )h ( n − k). By superposition, the total response to the series of the short pulses at time t is given by k=∞  y(t) = u(k )h (t − k). (3.4) k=0

If we take the limit as → 0, the basic pulse gets more and more narrow and taller and taller while holding a constant area. We then have the concept of an impulse signal, δ(t), and that will allow us to treat continuous signals. In that case we have, (3.5) lim p (t) = δ(t), →0

lim h (t) = h(t) = the impulse response.

→0

(3.6)

Moreover, in the limit as → 0, the summation in Eq. (3.4) is replaced by the integral

∞ u(τ )h(t − τ ) dτ , (3.7) y(t) = that is the convolution integral.

0

78 Chapter 3 Dynamic Response Figure 3.1 Illustration of convolution as the response of a system to a series of short pulse (impulse) input signals

System output h (t)

Input signal p (t)

1 ∆

1 0 ∆

t (sec)

0

(a)

t (sec)

h (t-t)

p (t-t)

1 ∆

1 0

t t+∆ t (sec)

0

t

t (sec)

(b) p (t)

y (t)

1 ∆

0 ∆ t t+∆

t (sec)

(c) p (t)

y (t)

1 ∆

0 ∆ t t+∆ 2t 2t+∆ 3t 3t+∆ t (sec) (d)

Impulse response Definition of impulse

1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 6 7 8 9 10 Time (sec)

The idea for the impulse comes from dynamics. Suppose we wish to study the motion of a baseball hit by a bat. The details of the collision between the bat and ball can be very complex as the ball deforms and the bat bends; however, for purposes of computing the path of the ball, we can summarize the effect of the collision as the net velocity change of the ball over a very short time period. We assume that the ball is subjected to an impulse, a very intense force for a very short time. The physicist Paul Dirac suggested that such forces could be represented by the mathematical concept of an impulse δ(t), which has the property that

δ(t) = 0 ∞

−∞

Sifting property of impulse

1.4 1.2 1 0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 6 7 8 9 10 Time (sec)

t = 0,

δ(t)dt = 1.

If f (t) is continuous at t = τ , then it has the “sifting property.”

(3.8) (3.9)

3.1 Review of Laplace Transforms Figure 3.2 Illustration of the representation of a general input signal as the sum of short pulses

79

u (t)

0 ∆ 2∆ 3∆

5∆

10∆



∞ −∞

t (sec)

f (τ )δ(t − τ ) dτ = f (t).

(3.10)

In other words, the impulse is so short and so intense that no value of f matters except over the short range where the δ occurs. Since integration is a limit of a summation process, Eq. (3.10) can be viewed as representing the function f as a sum of impulses. If we replace f by u, then Eq. (3.10) represents an input u(t) as a sum of impulses of intensity u(t − τ ). To find the response to an arbitrary input, the principle of superposition tells us that we need only find the response to a unit impulse. If the system is not only linear but also time invariant (LTI), then the impulse response is given by h(t −τ ) because the response at t to an input applied at τ depends only on the difference between the time the impulse is applied and the time we are observing the response, i.e. the elapsed time. Time invariant systems are called shift invariant for this reason. For time invariant systems, the output for a general input is given by the integral

∞ y(t) = u(τ )h(t − τ ) dτ , (3.11) −∞

or by changing of variables as τ1 = t − τ

y(t) =

−∞ ∞

u(t − τ1 )h(τ1 ) (−dτ1 ) =

The convolution integral

This is the convolution integral.

EXAMPLE 3.3

Convolution

∞ −∞

h(τ )u(t − τ ) dτ .

(3.12)

We can illustrate convolution with a simple system. Determine the impulse response for the system described by the differential equation y˙ + ky = u = δ(t), with an initial condition of y(0) = 0 before the impulse.

80 Chapter 3 Dynamic Response Solution. Because δ(t) has an effect only near t = 0, we can integrate this equation from just before zero to just after zero with the result that

0+

0+

0+ y˙ dt + k y dt = δ(t) dt. 0−

0−

0−

The integral of y˙ is simply y, the integral of y over so small a range is zero, and the integral of the impulse over the same range is unity. Therefore, y(0+ ) − y(0− ) = 1. Because the system was at rest before application of the impulse, y(0− ) = 0. Thus the effect of the impulse is that y(0+ ) = 1. For positive time we have the differential equation y˙ + ky = 0, y(0+ ) = 1. If we assume a solution y = Aest , then y˙ = Asest . The preceding equation then becomes Asest + kAest = 0, s + k = 0, s = −k. y(0+ )

Unit step

= 1, it is necessary that A = 1. Thus the solution for the impulse Because response is y(t) = h(t) = e−kt for t > 0. To take care of the fact that h(t) = 0 for negative time, we define the unit-step function  0, t < 0, 1(t) = 1, t ≥ 0. With this definition, the impulse response of the first-order system becomes h(t) = e−kt 1(t). The response of this system to a general input is given by the convolution of this impulse response with the input:

∞ h(τ )u(t − τ ) dτ y(t) =

−∞ ∞



−∞ ∞

= =

e−kτ 1(τ )u(t − τ ) dτ e−kτ u(t − τ ) dτ .

0

3.1.2 Transfer Functions and Frequency Response An immediate consequence of convolution is that an input of the form est results in an output H(s)est . Note that both input and output are exponential time functions, and that the output differs from the input only in the amplitude H(s). H(s) is the transfer function of the system. The constant s may be complex, expressed as s = σ + jω.

3.1 Review of Laplace Transforms

81

Thus, both the input and the output may be complex. If we let u(t) = est in Eq. (3.12), then

∞ y(t) = h(τ )u(t − τ ) dτ −∞ ∞

=

−∞ ∞

= =

−∞

∞ −∞

h(τ )es(t−τ ) dτ h(τ )est e−sτ dτ h(τ )e−sτ dτ est

= H(s)est , where1

H(s) =

Transfer function

∞ −∞

h(τ )e−sτ dτ .

(3.13) (3.14)

The integral in Eq. (3.14) does not need to be computed to find the transfer function of a system. Instead, one can assume a solution of the form of Eq. (3.13), substitute that into the differential equation of the system, then solve for the transfer function H(s). The transfer function can be formally defined as follows: The function H(s), which is the transfer gain from U(s) to Y (s)—input to output—is called the transfer function of the system. It is the ratio of the Laplace transform of the output to the Laplace transform of the input, Y (s) = H(s), (3.15) U(s) with the key assumption that all of the initial conditions on the system are zero. If the input u(t) is the unit impulse δ(t), then y(t) is the unit impulse response. The Laplace transform of u(t) is 1 and the transform of y(t) is H(s) because Y (s) = H(s).

(3.16)

In words, this is to say Transfer function

The transfer function H(s) is the Laplace transform of the unit impulse response h(t). Thus if one wishes to characterize an LTI system, one applies a unit impulse and the resulting response is a description (the inverse Laplace transform) of the transfer function.

EXAMPLE 3.4

Transfer Function Compute the transfer function for the system of Example 3.1, and find the output y for the input u = est . 1 Notice that this input is exponential for all time and Eq. (3.14) represents the response for all time. If the  system is causal, then h(t) = 0 for t < 0, and the integral reduces to H(s) = 0∞ h(τ )e−sτ dτ .

82 Chapter 3 Dynamic Response Solution. The system equation from Example 3.3 is y˙ (t) + ky(t) = u(t) = est . We assume that we can express y(t) as H(s)est . With this form, we have y˙

(3.17) = sH(s)est ,

and Eq. (3.17) reduces to sH(s)est + kH(s)est = est .

(3.18)

Solving for the transfer function H(s), we get 1 . H(s) = s+k Substituting this back into Eq. (3.13) yields the output est . y= s+k

Frequency response

A very common way to use the exponential response of LTIs is in finding the frequency response, or response to a sinusoid. First we express the sinusoid as a sum of two exponential expressions (Euler’s relation): A A cos(ωt) = (e jωt + e−jωt ). 2 If we let s = jω in the basic response formula Eq. (3.13), then the response to u(t) = ejωt is y(t) = H(jω)ejωt ; similarly, the response to u(t) = e−jωt is H(−jω)e−jωt . By superposition, the response to the sum of these two exponentials, which make up the cosine signal, is the sum of the responses: A (3.19) y(t) = [H(jω)e jωt + H(−jω)e−jωt ]. 2 The transfer function H(jω) is a complex number that can be represented in polar form or in magnitude-and-phase form as H( jω) = M(ω)e jϕ(ω) , or simply H = Me jϕ . With this substitution, Eq. (3.19) becomes  A
y(t) = M e j(ωt+ϕ) + e−j(ωt+ϕ) 2 = AM cos(ωt + ϕ), (3.20) where M = |H( jω)|, ϕ = ∠H( jω). This means that if a system represented by the transfer function H(s) has a sinusoidal input with magnitude A, the output will be sinusoidal at the same frequency with magnitude AM and will be shifted in phase by the angle ϕ.

EXAMPLE 3.5

Frequency Response For the system in Example 3.1, find the response to the sinusoidal input u = A cos(ωt). That is, a. find the frequency response and plot the response for k = 1, b. determine the complete response due to the sinusoidal input u(t) = sin(10t) again with k = 1.

3.1 Review of Laplace Transforms

83

Solution. In Example 3.4 we found the transfer function. To find the frequency response, we let s = jω so that 1 1 =⇒ H(jω) = . s+k jω + k

H(s) = From this we get M=√

1 ω2 + k 2

and ϕ = − tan−1


ω k

.

Therefore, the response of this system to a sinusoid will be y(t) = AM cos(ωt + ϕ).

(3.21)

M is usually referred to as the amplitude ratio and ϕ is referred to as the phase and they are both functions of the input frequency, ω.The MATLAB
program that follows is used to compute the amplitude ratio and phase for k = 1, as shown in Fig. 3.3. The logspace command is used to set the frequency range (on a logarithmic scale) and the bode command is used to compute the frequency response in MATLAB. Presenting frequency response in this manner (i.e., on a log–log scale) was originated by H. W. Bode; thus, these plots are referred to as “Bode plots.”2 (See Chapter 6, Section 6.1.) k = 1; numH = 1; denH = [1 k]; Figure 3.3 Frequency response for k=1

% form numerator % form denominator Magnitude

M

100

10⫺1

10⫺2 10⫺2

10⫺1 v

100 (rad/sec)

101

102

101

102

Phase

0

f⬚

⫺30 ⫺60 ⫺90 10⫺2

10⫺1 v

100 (rad/sec)

2 Note that % is used in MATLAB to denote comments.

84 Chapter 3 Dynamic Response sysH = tf(numH,denH);

% define system by its numerator and denominator % set frequency w to 50 values from 10−2 to 10+2 % compute frequency response % log–log plot of magnitude % semi log plot of phase

w = logspace(−2,2); [mag,phase] = bode(sysH,w); loglog(w,squeeze(mag)); semilogx(w,squeeze(phase));

To determine the response to an input that begins at t = 0 as u(t) = sin(10t)1(t), notice that from Laplace transform tables (Appendix A, Table A.2), we have 10 , + 100 where L denotes the Laplace transform, and the output of the system using partial fraction expansion (see Section 3.1.5) is given by L{u(t)} = L{sin(10t)} =

s2

Y (s) = H(s)U(s) 1 10 = 2 s + 1 s + 100 α0∗ α1 α0 + = + s + 1 s + j10 s − j10 =

10 101

s+1

+

j 2(1−j10)

s + j10

+

−j 2(1+j10)

s − j10

.

The inverse Laplace transform of the output is given by (see Appendix A) 10 −t 1 e +√ sin(10t + ϕ) 101 101 = y1 (t) + y2 (t),

y(t) =

where ϕ = tan−1 (−10) = −84.2◦ . The component y1 (t) is called the transient response as it decays to zero as time goes on and the component y2 (t) is called the steady state and equals the response given by Eq. (3.21). Figure 3.4(a) is a plot of the time history of the output showing the different components (y1 , y2 ) and the composite (y) output response. The output frequency is 10 rad/sec and the steady-state phase difference measured from Fig. 3.4(b) is approximately 10*δt = 1.47 rad = 84.2◦3 . Figure 3.4(b) shows the output lags the input by 84.2◦ . Figure 3.4(b) shows that the steady-state amplitude of the output is the amplitude ratio √ 1 = 0.0995 (i.e., the amplitude of the input signal times the 101 magnitude of the transfer function evaluated at ω = 10 rad/sec). This example illustrates that the response of an LTI system to a sinusoid of frequency ω is a sinusoid with the same frequency and with an amplitude ratio equal 3 The phase difference may also be determined by a Lissajous pattern.

3.1 Review of Laplace Transforms Figure 3.4 (a) Complete transient response; (b) phase lag between output and input

85

0.25 Transient response 0.2

y2

y

0.15 y1

Output

0.1 0.05 0 ⫺0.05 ⫺0.1

0

1

2

3

4

5 6 Time (sec) (a)

7

8

9

10

9.7

9.8

9.9

10

1 Steady-state response 0.8 u(t) 0.6

Output, input

0.4 0.2 y(t) 0 ⫺0.2

dt

⫺0.4 ⫺0.6 ⫺0.8 ⫺1

9

9.1

9.2

9.3

9.4

9.5 9.6 Time (sec) (b)

to the magnitude of the transfer function evaluated at the input frequency. Furthermore, the phase difference between input and output signals is given by the phase of the transfer function evaluated at the input frequency. The magnitude ratio and phase difference can be computed from the transfer function as just discussed; they can also be measured experimentally quite easily in the laboratory by driving the system with a known sinusoidal input and measuring the steady-state amplitude and phase of the system’s output. The input frequency is set to sufficiently many values so that curves such as the one in Fig. 3.3 are obtained.

86 Chapter 3 Dynamic Response We can generalize the frequency response by defining the Laplace transform of a signal f (t) as

∞ F(s) = f (t)e−sτ dt. (3.22) −∞

The key property of Laplace transforms

If we apply this definition to both u(t) and y(t) and use the convolution integral Eq. (3.12), we find that Y (s) = H(s)U(s),

Transient response

(3.23)

where Y (s) and U(s) are the Laplace transforms of y(t) and u(t), respectively. We prove this result in Appendix A. Laplace transforms such as Eq. (3.22) can be used to study the complete response characteristics of feedback systems, including the transient response—that is, the time response to an initial condition or suddenly applied signal. This is in contrast to the use of Fourier transforms, where the steady-state response is the main concern. A standard problem in control is to find the response y(t) of a system given the input u(t) and a model of the system. With Eq. (3.22) we have a means for computing the response of LTI systems to quite general inputs. Given any input into a system, we compute the transform of the input and the transfer function for the system. The transform of the output is then given by Eq. (3.23) as the product of these two. If we wanted the time function of the output, we would need to “invert” Y (s) to get what is called the inverse transform; this step is typically not carried out explicitly. Nevertheless, understanding the process necessary for deriving y(t) from Y (s) is important because it leads to insight into the behavior of linear systems. Hence, given a general linear system with transfer function H(s) and an input signal u(t), the procedure for determining y(t) using the Laplace transform is given by the following steps: STEP 1. Determine the transfer function: H(s) = L{impulse response of the system}. Compute H(s) by the following steps: (a) Take the Laplace transform of the equations of motion. A table of transform properties is frequently useful in this process. (b) Solve the resulting algebraic equations. Often this step is greatly helped by drawing the corresponding block diagram and solving the equations by graphical manipulation of the blocks or using MATLAB. STEP 2. Determine the Laplace transform of the input signal: U(s) = L{u(t)}. STEP 3. Determine the Laplace transform of the output: Y (s) = H(s)U(s). STEP 4. Break up Y (s) by partial-fraction expansion. STEP 5. Find the output of the system by computing the inverse Laplace transform of Y (s) in Step 4, y(t) = L−1 {Y (s)} [i.e., invert Y (s) to get y(t)]: (a) Look up the components of y(t) in a table of transform–time function pairs. (b) Combine the components to give the total solution in the desired form. As already mentioned, Steps 4 and 5 are almost never carried out in practice, and a modified solution for a qualitative rather than a quantitative solution is often adequate and almost always used for control design purposes. The process begins with the first three steps as before. However, rather than inverting Y (s), one can use

3.1 Review of Laplace Transforms

87

prior knowledge and intuition about the effects of pole and zero locations in Y (s) on the response y(t) to estimate key parameters of y(t). That is, we get information about y(t) from the pole–zero constellation of Y (s) without actually inverting it, as discussed in the rest of this chapter. While it is possible to determine the transient response properties of the system using Eq. (3.22), it is generally more useful to use a simpler version of the Laplace transform based on the input beginning at time zero.

3.1.3 The L− Laplace Transform

Definition of Laplace transform

In many applications it is useful to define a one-sided (or unilateral) Laplace transform, which uses 0− (that is, a value just before t = 0) as the lower limit of integration in Eq. (3.22). The L− Laplace transform of f (t), denoted by L− {f (t)} = F(s), is a function of the complex variable s = σ + jω, where

∞ F(s) = f (t)e−st dt. (3.24) 0−

The decaying exponential term in the integrand in effect provides a built-in convergence factor if σ > 0. This means that even if f (t) does not vanish as t → ∞, the integrand will vanish for sufficiently large values of σ if f does not grow at a fasterthan-exponential rate. The fact that the lower limit of integration is at 0− allows the use of an impulse function at t = 0, as illustrated in Example 3.3; however, this distinction between t = 0− and t = 0 does not usually come up in practice. We will therefore for the most part drop the minus superscript on t = 0; however, we will return to using the notation t = 0− when an impulse at t = 0 is involved and the distinction is of practical value. If Eq. (3.24) is a one-sided transform, then by extension, Eq. (3.22) is a two-sided Laplace transform.4 We will use the L symbol from here on to mean L− . On the basis of the formal definition in Eq. (3.24), we can ascertain the properties of Laplace transforms and compute the transforms of common time functions. The analysis of linear systems by means of Laplace transforms usually involves using tables of common properties and time functions, so we have provided this information in Appendix A. The tables of time functions and their Laplace transforms, together with the table of properties, permit us to find transforms of complex signals from simpler ones. For a thorough study of Laplace transforms and extensive tables, see Churchill (1972) and Campbell and Foster (1948). For more study of the two-sided transform, see Van der Pol and Bremmer (1955). These authors show that the time function can be obtained from the Laplace transform by the inverse relation

σc +j∞ 1 f (t) = F(s)est ds, (3.25) 2πj σc −j∞ where σc is a selected value to the right of all the singularities of F(s) in the s-plane. In practice, this relation is seldom used. Instead, complex Laplace transforms are broken down into simpler ones that are listed in the tables along with their corresponding time responses. 4 The other possible one-sided transform is, of course, L , in which the lower limit of the integral is 0+ . + It is sometimes used in other applications.

88 Chapter 3 Dynamic Response Let us compute a few Laplace transforms of some typical time functions.

EXAMPLE 3.6

Step and Ramp Transforms Find the Laplace transform of the step a1(t) and ramp bt1(t) functions. Solution. For a step of size a, f (t) = a1(t), and from Eq. (3.24) we have ∞

∞ a −ae−st  −a −st = , Re(s) > 0. F(s) = ae dt = =0−  s s s 0 0 For the ramp signal f (t) = bt1(t), again from Eq. (3.24) we have ∞ 

∞ be−st bte−st b −st F(s) = − 2 bte dt = − = 2, Re(s) > 0, s s s 0 0 where we employed the technique of integration by parts,



u dv = uv − v du, with u = bt and dv = e−st dt. We can then extend the domain of the validity of F(s) to the entire s-plane except at the pole location namely the origin (see Appendix A). A more subtle example is that of the impulse function.

EXAMPLE 3.7

Impulse Function Transform Find the Laplace transform of the unit-impulse function. Solution. From Eq. (3.24) we get

0+

∞ δ(t)e−st dt = δ(t) dt = 1. F(s) = 0−

0−

(3.26)

It is the transform of the unit-impulse function that led us to choose the L− transform rather than the L+ transform.

EXAMPLE 3.8

Sinusoid Transform Find the Laplace transform of the sinusoid function. Solution. Again, we use Eq. (3.24) to get

∞ (sin ωt)e−st dt. L{sin ωt} = 0

If we substitute the relation from Eq. (D.34) in Appendix WD, sin ωt =

e jωt − e−jωt 2j

(3.27)

3.1 Review of Laplace Transforms into Eq. (3.27), we find that

L{sin ωt} =

∞  e jωt 0

=

1 2j



∞


− e−jωt 2j



89

e−st dt

 e( jω−s)t − e−(jω+s)t dt

0

∞  1 1 ( jω−s)t −( jω+s)t  e e −  jω − s jω + s 0 ω = 2 , Re(s) > 0. s + ω2 We can then extend the domain of the validity of computed Laplace transform to the entire s-plane except at the pole locations s = ±jω (see Appendix A). 1 = 2j



Table A.2 in Appendix A lists Laplace transforms for elementary time functions. Each entry in the table follows from direct application of the transform definition of Eq. (3.24), as demonstrated by Examples 3.6 to 3.8.

3.1.4 Properties of Laplace Transforms In this section we will address each of the significant properties of the Laplace transform listed in Table A.1. For the proofs of these properties and related examples as well as the Initial Value Theorem, the reader is referred to Appendix A.

1. Superposition One of the more important properties of the Laplace transform is that it is linear: L{αf1 (t) + βf2 (t)} = αF1 (s) + βF2 (s).

(3.28)

The amplitude scaling property is a special case of this; that is, L{αf (t)} = αF(s).

(3.29)

2. Time Delay Suppose a function f (t) is delayed by λ > 0 units of time. Its Laplace transform is

∞ f (t − λ)e−st dt = e−sλ F(s). (3.30) F1 (s) = 0

From this result we see that a time delay of λ corresponds to multiplication of the transform by e−sλ .

3. Time Scaling It is sometimes useful to time-scale equations of motion. For example, in the control system of a disk drive, it is meaningful to measure time in milliseconds (see also Chapter 10). If the time t is scaled by a factor a, then the Laplace transform of the time-scaled signal is

∞ 1
s F . (3.31) f (at)e−st dt = F1 (s) = |a| a 0

90 Chapter 3 Dynamic Response 4. Shift in Frequency Multiplication (modulation) of f (t) by an exponential expression in the time domain corresponds to a shift in frequency:

∞ e−at f (t)e−st dt = F(s + a). (3.32) F1 (s) = 0

5. Differentiation The transform of the derivative of a signal is related to its Laplace transform and its initial condition as  follows:  ∞  df df = e−st dt = −f (0− ) + sF(s). (3.33) L − dt dt 0 Another application of Eq. (3.33) leads to (3.34) L{f¨ } = s2 F(s) − sf (0− ) − f˙ (0− ). Repeated application of Eq. (3.33) leads to L{f m (t)} = sm F(s) − sm−1 f (0− ) − sm−2 f˙ (0− ) − · · · − f (m−1) (0− ), where

f m (t)

(3.35)

denotes the mth derivative of f (t) with respect to time.

6. Integration Let us assume that we wish to determine the Laplace transform of the integral of a time function f (t); that is,  t  1 (3.36) F1 (s) = L f (ξ ) dξ = F(s), s 0 which means that we simply multiply the function’s Laplace transform by 1s .

7. Convolution We have seen previously that the response of a system is determined by convolving the input with the impulse response of the system, or by forming the product of the transfer function and the Laplace transform of the input. The discussion that follows extends this concept to various time functions. Convolution in the time domain corresponds to multiplication in the frequency domain. Assume that L{f1 (t)} = F1 (s) and L{f2 (t)} = F2 (s). Then

∞ f1 (t) ∗ f2 (t)e−st dt = F1 (s)F2 (s). (3.37) L{f1 (t) ∗ f2 (t)} = 0

This implies that

L−1 {F1 (s)F2 (s)} = f1 (t) ∗ f2 (t). A similar, or dual, of this result is discussed next.

(3.38)

8. Time Product Multiplication in the time domain corresponds to convolution in the frequency domain: 1 F1 (s) ∗ F2 (s). (3.39) L{f1 (t)f2 (t)} = 2πj

3.1 Review of Laplace Transforms

91

9. Multiplication by Time Multiplication by time corresponds to differentiation in the frequency domain: L{tf (t)} = −

d F(s). ds

(3.40)

3.1.5 Inverse Laplace Transform by Partial-Fraction Expansion The easiest way to find f (t) from its Laplace transform F(s), if F(s) is rational, is to expand F(s) as a sum of simpler terms that can be found in the tables. The basic tool for performing this operation is called partial-fraction expansion. Consider the general form for the rational function F(s) consisting of the ratio of two polynomials: F(s) =

b1 sm + b2 sm−1 + · · · + bm+1 . sn + a1 sn−1 + · · · + an

(3.41)

By factoring the polynomials, this same function could also be expressed in terms of the product of factors as m (s − zi ) F(s) = K ni=1 . (3.42) i=1 (s − pi )

Zeros and poles

We will discuss the simple case of distinct poles here. For a transform F(s) representing the response of any physical system, m ≤ n. When s = zi , s is referred to as a zero of the function, and when s = pi , s is referred to as a pole of the function. Assuming for now that the poles {pi } are real or complex but distinct, we rewrite F(s) as the partial fraction F(s) =

C1 C2 Cn + + ··· + . s − p1 s − p2 s − pn

(3.43)

Next, we determine the set of constants {Ci }. We multiply both sides of Eq. (3.43) by the factor s − p1 to get (s − p1 )F(s) = C1 +

s − p1 (s − p1 )Cn C2 + · · · + . s − p2 s − pn

(3.44)

If we let s = p1 on both sides of Eq. (3.44), then all the Ci terms will equal zero except for the first one. For this term, C1 = (s − p1 )F(s)|s=p1 .

(3.45)

The other coefficients can be expressed in a similar form: Ci = (s − pi )F(s)|s=pi . The cover-up method of determining coefficients

This process is called the cover-up method because, in the factored form of F(s) [Eq. (3.42)], we can cover up the individual denominator terms, evaluate the rest of the expression with s = pi , and determine the coefficients Ci . Once this has been completed, the time function becomes f (t) =

n  i=1

Ci epi t 1(t)

92 Chapter 3 Dynamic Response because, as entry 7 in Table A.2 shows, if F(s) =

1 , s − pi

then f (t) = epi t 1(t). For the cases of quadratic factors or repeated roots in the denominator, seeAppendixA.

EXAMPLE 3.9

Partial-Fraction Expansion: Distinct Real Roots Suppose you have computed Y (s) and found that (s + 2)(s + 4) . Y (s) = s(s + 1)(s + 3) Find y(t). Solution. We may write Y (s) in terms of its partial-fraction expansion: C2 C3 C1 + + . Y (s) = s s+1 s+3 Using the cover-up method, we get  (s + 2)(s + 4)  8 C1 = = .  (s + 1)(s + 3) 3 s=0

In a similar fashion,

 (s + 2)(s + 4)  3 =− C2 = s(s + 3) s=−1 2

 (s + 2)(s + 4)  1 =− . C3 =  s(s + 1) 6 s=−3 We can check the correctness of the result by adding the components again to verify that the original function has been recovered. With the partial fraction the solution can be looked up in the tables at once to be 8 3 1 y(t) = 1(t) − e−t 1(t) − e−3t 1(t). 3 2 6 The partial fraction expansion may be computed using the residue function in MATLAB: and

num = conv([1 2],[1 4]); den = conv([1 1 0],[1 3]); [r,p,k] = residue(num,den);

% form numerator polynomial % form denominator polynomial % compute the residues

which yields the result r = [−0.1667 −1.5000 2.6667]
;

p = [−3 −1 0]
;

k = [ ];

and agrees with the hand calculations. Note that the conv function in MATLAB is used to multiply two polynomials. (The arguments of the function are the polynomial coefficients.)

3.1 Review of Laplace Transforms

93

3.1.6 The Final Value Theorem An especially useful property of the Laplace transform in control known as the Final Value Theorem allows us to compute the constant steady-state value of a time function given its Laplace transform. The theorem follows from the development of partialfraction expansion. Suppose we have a transform Y (s) of a signal y(t) and wish to know the final value y(t) from Y (s). There are three possibilities for the limit. It can be constant, undefined, or unbounded. If Y (s) has any poles (i.e., denominator roots, as described in Section 3.1.5) in the right half of the s-plane—that is, if the real part of any pi > 0—then y(t) will grow and the limit will be unbounded. If Y (s) has a pair of poles on the imaginary axis of the s-plane (i.e., pi = ±jω), then y(t) will contain a sinusoid that persists forever and the final value will not be defined. Only one case can provide a nonzero constant final value: If all poles of Y (s) are in the left half of the s-plane, except for one at s = 0, then all terms of y(t) will decay to zero except the term corresponding to the pole at s = 0, and that term corresponds to a constant in time. Thus, the final value is given by the coefficient associated with the pole at s = 0. Therefore, the Final Value Theorem is as follows: The Final Value Theorem

If all poles of sY (s) are in the left half of the s-plane, then lim y(t) = lim sY (s).

t→∞

s→0

(3.46)

This relationship is proved in Appendix A.

EXAMPLE 3.10

Final Value Theorem Find the final value of the system corresponding to Y (s) =

3(s + 2) . s(s2 + 2s + 10)

Solution. Applying the Final Value Theorem, we obtain 3·2 = 0.6. 10 Thus, after the transients have decayed to zero, y(t) will settle to a constant value of 0.6. y(∞) = sY (s)|s=0 =

Use the Final Value Theorem on stable systems only

Care must be taken to apply the Final Value Theorem only to stable systems (see Section 3.6). While one could use Eq. (3.46) on any Y (s), doing so could result in erroneous results, as shown in the next example.

EXAMPLE 3.11

Incorrect Use of the Final Value Theorem Find the final value of the signal corresponding to Y (s) =

3 . s(s − 2)

94 Chapter 3 Dynamic Response Solution. If we blindly apply Eq. (3.46), we obtain 3 y(∞) = sY (s)|s=0 = − . 2 However,   3 3 y(t) = − + e2t 1(t), 2 2 and Eq. (3.46) yields the constant term only. Of course, the true final value is unbounded. Calculating DC gain by the Final Value Theorem

The theorem can also be used to find the DC gain of a system. The DC gain is the ratio of the output of a system to its input (presumed constant) after all transients have decayed. To find the DC gain, we assume that there is a unit-step input [U(s) = 1/s] and we use the Final Value Theorem to compute the steady-state value of the output. Therefore, for a system transfer function G(s), 1 (3.47) DC gain = lim sG(s) = lim G(s). s→0 s→0 s

EXAMPLE 3.12

DC Gain Find the DC gain of the system whose transfer function is 3(s + 2) . G(s) = 2 (s + 2s + 10) Solution. Applying Eq. (3.47), we get 3·2 = 0.6. DC gain = G(s)|s=0 = 10

3.1.7 Using Laplace Transforms to Solve Problems Laplace transforms can be used to solve differential equations using the properties described in Appendix A. First, we find the Laplace transform of the differential equation using the differentiation properties in Eqs. (A.12) and (A.13) in Appendix A. Then we find the Laplace transform of the output; using partial-fraction expansion and Table A.2, this can be converted to a time response function. We will illustrate this with three examples.

EXAMPLE 3.13

Homogeneous Differential Equation Solution Find the solution to the differential equation y¨ (t) + y(t) = 0, where

y(0) = α, y˙ (0) = β.

Solution. Using Eq. (3.34), the Laplace transform of the differential equation is s2 Y (s) − αs − β + Y (s) = 0, (s2 + 1)Y (s) = αs + β, β αs + . Y (s) = 2 s + 1 s2 + 1

3.1 Review of Laplace Transforms

95

After looking up in the transform tables (Table A.2, Appendix A) the two terms on the right side of the preceding equation, we get y(t) = [α cos t + β sin t]1(t), where 1(t) denotes a unit step function. We can verify that this solution is correct by substituting it back into the differential equation. Another example will illustrate the solution when the equations are not homogeneous—that is, when the system is forced.

EXAMPLE 3.14

Forced Differential Equation Solution Find the solution to the differential equation y¨ (t)+5˙y(t)+4y(t) = 3, where y(0) = α, y˙ (0) = β. Solution. Taking the Laplace transform of both sides using Eqs. (3.33) and (3.34), we get 3 s2 Y (s) − sα − β + 5[sY (s) − α] + 4Y (s) = . s Solving for Y (s) yields Y (s) =

s(sα + β + 5α) + 3 . s(s + 1)(s + 4)

The partial-fraction expansion using the cover-up method is Y (s) =

3 4

−

3−β−4α 3

+

3−4α−4β 12

. s s+1 s+4 Therefore, the time function is given by   3 −3 + β + 4α −t 3 − 4α − 4β −4t 1(t). + e + e y(t) = 4 3 12 By differentiating this solution twice and substituting the result in the original differential equation, we can verify that this solution satisfies the differential equation. The solution is especially simple if the initial conditions are all zero.

EXAMPLE 3.15

Forced Equation Solution with Zero Initial Conditions Find the solution to y¨ (t)+5˙y(t)+4y(t) = u(t), y(0) = 0, y˙ (0) = 0, u(t) = 2e−2t 1(t), 1. using partial-fraction expansion and 2. using MATLAB. Solution 1. Taking the Laplace transform of both sides, we get s2 Y (s) + 5sY (s) + 4Y (s) =

2 . s+2

96 Chapter 3 Dynamic Response Solving for Y (s) yields 2 . (s + 2)(s + 1)(s + 4) The partial-fraction expansion using the cover-up method is Y (s) =

2 1 1 + 3 + 3 . s+2 s+1 s+4 Therefore, the time function is given by   2 1 y(t) = −1e−2t + e−t + e−4t 1(t). 3 3

Y (s) = −

2. The partial-fraction expansion may also be computed using the MATLAB residue function, num = 2; den = poly([−2;−1;−4]); [r,p,k] = residue(num,den);

% form numerator % form denominator polynomial from its roots % compute the residues

which results in the desired answer r = [0.3333 −1 0.6667]
;

p = [−4 −2 −1]
;

k = [ ];

and agrees with the hand calculations.

Poles indicate response character

The primary value of using the Laplace transform method of solving differential equations is that it provides information concerning the qualitative characteristic behavior of the response. Once we know the values of the poles of Y (s), we know what kind of characteristic terms will appear in the response. In the last example the pole at s = −1 produced a decaying y = Ce−t term in the response. The pole at s = −4 produced a y = Ce−4t term in the response, which decays faster. If there had been a pole at s = +1, there would have been a growing y = Ce+t term in the response. Using the pole locations to understand in essence how the system will respond is a powerful tool and will be developed further in Section 3.3. Control systems designers often manipulate design parameters so that the poles have values that would give acceptable responses, and they skip the steps associated with converting those poles to actual time responses until the final stages of the design. They use trial-and-error design methods (as described in Chapter 5) that graphically present how changes in design parameters affect the pole locations. Once a design has been obtained, with pole locations predicted to give acceptable responses, the control designer determines a time response to verify that the design is satisfactory. This is typically done by computer, which solves the differential equations directly by using numerical computer methods.

3.1.8 Poles and Zeros A rational transfer function can be described either as a ratio of two polynomials in s, H(s) =

N(s) b1 sm + b2 sm−1 + · · · + bm+1 , = sn + a1 sn−1 + · · · + an D(s)

(3.48)

3.1 Review of Laplace Transforms

97

or as a ratio in factored zero pole form

m (s − zi ) H(s) = K ni=1 . i=1 (s − pi )

Zeros

(3.49)

K is called the transfer function gain. The roots of the numerator z1, z2,... , zm are called the finite zeros of the system. The zeros are locations in the s-plane where the transfer function is zero. If s = zi , then H(s)|s=zi = 0.

Poles

The zeros also correspond to the signal transmission-blocking properties of the system and are also called the transmission zeros of the system. The system has the inherent capability to block frequencies coinciding with its zero locations. If we excite the system with the nonzero input, u = u0 es0 t , where s0 is not a pole of the system, then the output is identically zero,5 y ≡ 0, for frequencies where s0 = zi . The zeros also have a significant effect on the transient properties of the system (see Section 3.5). The roots of the denominator, p1, p2, . . . , pn are called the poles6 of the system. The poles are locations in the s-plane where the magnitude of the transfer function becomes infinite. If s = pi , then |H(s)|s=pi = ∞. The poles of the system determine its stability properties, as we shall see in Section 3.6. The poles of the system also determine the natural or unforced behavior of the system, referred to as the modes of the system. The zeros and poles may be complex quantities, and we may display their locations in a complex plane, which we refer to as the s-plane. The locations of the poles and zeros lie at the heart of feedback control design and have significant practical implications for control system design. The system is said to have n−m zeros at infinity if m < n because the transfer function approaches zero as s approaches infinity. If the zeros at infinity are also counted, the system will have the same number of poles and zeros. No physical system can have n < m; otherwise, it would have an infinite response at ω = ∞. If zi = pj , then there are cancellations in the transfer function, which may lead to undesirable system properties as discussed in Chapter 7.

3.1.9 Linear System Analysis Using MATLAB The first step in analyzing a system is to write down (or generate) the set of timedomain differential equations representing the dynamic behavior of the physical system. These equations are generated from the physical laws governing the system behavior—for example, rigid body dynamics, thermo-fluid mechanics, and electromechanics, as described in Chapter 2. The next step in system analysis is to determine and designate inputs and outputs of the system and then to compute the 5 Identically zero means that the output and all of its derivatives are zero for t > 0. 6 The meaning of the pole can also be appreciated by visualizing a 3-D plot of the transfer function, where

the real and imaginary parts of s are plotted on the x and y axes, and the magnitude of the transfer function is plotted on the vertical z axis. For a single pole, the resulting 3-D plot will look like a tent with the “tent-pole” being located at the pole of the transfer function!

98 Chapter 3 Dynamic Response transfer function characterizing the input–output behavior of the dynamic system. Earlier in this chapter we saw that a linear dynamic system may also be represented by the Laplace transform of its differential equation—that is, its transfer function. The transfer function may be expressed as a ratio of two polynomials as in Eq. (3.48) or in factored zero–pole form as in Eq. (3.49). By analyzing the transfer function, we can determine the dynamic properties of the system, both in a qualitative and quantitative manner. One way of extracting useful system information is simply to determine the pole–zero locations and deduce the essential characteristics of the dynamic properties of the system. Another way is to determine the time-domain properties of the system by determining the response of the system to typical excitation signals such as impulses, steps, ramps, and sinusoids. Yet another way is to determine the time response analytically by computing the inverse Laplace transform using partial-fraction expansions and Tables A.1 and A.2. Of course, it is also possible to determine the system response to an arbitrary input. We will now illustrate this type of analysis by carrying out the preceding calculations for some of the physical systems addressed in the examples in Chapter 2 in order of increasing degree of difficulty. We will go back and forth between the different representations of the system, transfer function, and pole–zero, etc., using MATLAB as our computational engine. MATLAB typically accepts the specification of a system in several forms, including transfer function and zero–pole, and refers to these two descriptions as tf and zp, respectively. Furthermore, it can transform the system description from any one form to another.

EXAMPLE 3.16

MATLAB printsys

EXAMPLE 3.17

Cruise Control Transfer Function Using MATLAB Find the transfer function between the input u and the position of the car x in the cruise control system in Example 2.1. Solution. From Example 2.1 we find that the transfer function of the system is 0s2 + 0s + 0.001 0.001 = . H(s) = 2 s + 0.05s + 0 s(s + 0.05) In MATLAB, the coefficients of the numerator polynomial are displayed as the row vector num and the denominator coefficients are displayed as den. The results for this example are num = [0 0 0.001] and den = [1 0.05 0]. They can be returned by MATLAB in this form using the printsys(num,den) command. The pole–zero description is computed using the MATLAB command [z, p, k] = tf2zp(num, den) and in the transfer function in factored form, where z = [ ], p =  would result
0 − 0.05 , and k = 0.001.

DC Motor Transfer Function Using MATLAB In Example 2.13, assume that Jm = 0.01 kg·m2 , b = 0.001 N·m·sec, Kt = Ke = 1, Ra = 10 , and La = 1 H. Find the transfer function between the input va and

3.1 Review of Laplace Transforms

99

1. the output θm , 2. the output ω = θ˙m . Solution 1. Substituting the preceding parameters into Example 2.13, we find that the transfer function of the system is H(s) =

s3

100 . + 10.1s2 + 101s

In MATLAB we display the coefficients of the numerator polynomial as the row vector numa and the denominator as dena. The results for this example are numa = [ 0

0

0

100 ]

and

dena = [ 1

10.1

0 ].

101

The pole–zero description is computed using the MATLAB command [z, p, k] = tf2zp(numa, dena) which results in  z = [ ], p = 0

−5.0500

+8.6889j

−5.0500

−8.6889j




,

k = 100,

and yields the transfer function in factored form: H(s) =

100 . s(s + 5.05 + j8.6889)(s + 5.05 − j8.6889)

2. If we consider the velocity θ˙m as the output, then we find numb=[0 0 100], denb=[1 10.1 101], which tells us that the transfer function is G(s) =

100 100s = 2 . s3 + 10.1s2 + 101s s + 10.1s + 101

This is as expected, because θ˙m is simply the derivative of θm ; thus L{θ˙m } = sL{θm }. For a unit step command in va , we can compute the step response in MATLAB (recall Example 2.1): numb=[0 0 100]; denb=[1 10.1 101]; sysb=tf(numb,denb); t=0:0.01:5; y=step(sysb,t) plot(t,y)

% form numerator % form denominator % define system by its numerator and denominator % form time vector % compute step response; % plot step response

The system yields a steady-state constant angular velocity as shown in Fig. 3.5. Note that there is a slight offset, since the system does not have unity DC gain.

When a dynamic system is represented by a single differential equation of any order, finding the polynomial form of the transfer function from that differential equation is usually easy. Therefore, you will find it best in these cases to specify a system directly in terms of its transfer function.

100 Chapter 3 Dynamic Response Figure 3.5 Transient response for DC motor

1.4 1.2

v (rad/sec)

0 0.8 0.6 0.4 0.2 0

EXAMPLE 3.18

0

1

2 3 Time (sec)

4

5

Transformations Using MATLAB Find the transfer function of the system whose differential equation is y¨ + 6˙y + 25y = 9u + 3˙u. Solution. Using the differentiation rules given by Eqs. (3.33) and (3.34), we see by inspection that 3s + 9 Y (s) = 2 . U(s) s + 6s + 25 The MATLAB statements are numG = [3 9]; denG = [1 6 25];

% form numerator % form denominator

If the transfer function was desired in factored form, it could be obtained by transforming the tf description. Therefore, the MATLAB statement % convert from numerator-denominator polynomials to pole–zero form [z,p,k] = tf2zp(numG,denG) would result in z = −3, p = [−3 + 4j − 3 − 4j]
, k = 3. This means that the transfer function could also be written as 3(s + 3) Y (s) = . U(s) (s + 3 − 4j)(s + 3 + 4j) We may also convert from zero–pole representation to the transfer function representation using the MATLAB zp2tf command % convert from pole–zero form to numerator-denominator polynomials [numG,denG]=zp2tf(z,p,k) For this example, z=[−3], p=[−3+i*4;−3−i*4], k=[3] will yield the numerator and denominator polynomials.

3.1 Review of Laplace Transforms

EXAMPLE 3.19

101

Satellite Transfer Function Using MATLAB 1. Find the transfer function between the input Fc and the satellite attitude θ in Example 2.3 and 2. Determine the response of the system to a 25-N pulse of 0.1 sec duration, starting at t = 5 sec. Let d = 1 m and I = 5000 kg-m2 . Solution 1 = 0.0002 1. From Example 2.3, dI = 5000 function of the system is



m kg-m2

and this means that the transfer

0.0002 , s2 which can also be determined by inspection for this particular case. We may display the coefficients of the numerator polynomial as the row vector num and the denominator as the row vector den. The results for this example are H(s) =

numG = [0

0

0.0002] and

denG = [1

0

0].

2. The following MATLAB statements compute the response of the system to a 25-N, 0.1-sec duration thrust pulse input: numG=[0 0 0.0002]; denG=[1 0 0]; sysG=tf(numG,denG); t=0:0.01:10; % pulse of 25N, at 5 sec, for 0.1 sec duration u1=[zeros(1,500) 25∗ones(1,10) zeros(1,491)]; [y1]=lsim(sysG,u1,t); ff=180/pi; y1=ff∗y1; plot(t,u1); plot(t,y1);

% form the transfer function % define system by its transfer function % set up time vector with dt = 0.01 sec

% pulse input % linear simulation % conversion factor from radians to degrees % output in degrees % plot input signal % plot output response

The system is excited with a short pulse (an impulsive input) that has the effect of imparting a nonzero angle θ0 at time t = 5 sec on the system. Because the system is undamped, in the absence of any control it drifts with constant angular velocity with a value imparted by the impulse at t = 5 sec. The time response of the input is shown in Fig. 3.6(a) along with the drift in angle θ in Fig. 3.6(b). We now excite the system with the same positive-magnitude thrust pulse at time t = 5 sec but follow that with a negative pulse with the same magnitude and duration at time t = 6.1 sec. [See Figure 3.7(a) for the input thrust.] Then the attitude response of the system is as shown in Figure 3.7(b). This is actually how the satellite attitude angle is controlled in practice. The additional relevant MATLAB statements are

102 Chapter 3 Dynamic Response 25

Thrust Fc

20 15 10 5 0

0

1

2

3

4 5 6 Time (sec) (a)

7

8

9

10

0

1

2

3

4 5 6 Time (sec) (b)

7

8

9

10

0.16 0.14 0.12 0.1 u


Figure 3.6 Transient response for satellite: (a) thrust input; (b) satellite attitude

0.08 0.06 0.04 0.02 0

% double pulse input u2=[zeros(1,500) 25∗ones(1,10) zeros(1,100) −25∗ones(1,10) zeros(1,381)]; [y2]=lsim(sysG,u2,t); % linear simulation plot(t,u2); % plot input signal ff=180/pi; % conversion factor from radians to degrees y2=ff∗y2; % output in degrees plot(t,y2); % plot output response

3.2 System Modeling Diagrams 3.2.1 The Block Diagram To obtain the transfer function, we need to find the Laplace transform of the equations of motion and solve the resulting algebraic equations for the relationship between the input and the output. In many control systems the system equations can be written so that their components do not interact except by having the input of one part be the

3.2 System Modeling Diagrams 25 20 15

Thrust, Fc

10 5 0 5 10 15 20 25

0

1

2

3

4 5 6 Time (sec)

7

8

9

10

7

8

9

10

(a) 0.035 0.03 0.025 0.02 u


Figure 3.7 Transient response for satellite (doublepulse): (a) thrust input; (b) satellite attitude

103

0.015 0.01 0.005 0

0

1

2

3

4 5 6 Time (sec) (b)

output of another part. In these cases, it is easy to draw a block diagram that represents the mathematical relationships in a manner similar to that used for the component block diagram in Fig. 1.2, Chapter 1. The transfer function of each component is placed in a box, and the input–output relationships between components are indicated by lines and arrows. We can then solve the equations by graphical simplification, which is often easier and more informative than algebraic manipulation, even though the methods are in every way equivalent. Drawings of three elementary block diagrams are seen in Fig. 3.8. It is convenient to think of each block as representing an electronic amplifier with the transfer function printed inside. The interconnections of blocks include summing points, where any number of signals may be added together. These are represented by a circle with the symbol
inside. In Fig. 3.8(a) the block with transfer function G1 (s) is in series with the block with transfer function G2 (s), and the overall transfer function is given by the product G2 G1 . In Fig. 3.8(b) two systems are in parallel with their outputs added, and the overall transfer function is given by the sum G1 + G2 . These diagrams derive simply from the equations that describe them.

104 Chapter 3 Dynamic Response R(s) U1(s)

Y2(s) G1

U(s)

G1





G2 G2

Y2(s)  G1G2 U1(s) (a)

Y(s)

 



U1(s)

 Y2(s)

G1

G2

Y1(s)

Y(s)

U2(s)

Y(s)  G1  G2 U(s)

G1 Y(s)  1  G1G2 R(s)

(b)

(c)

Figure 3.8 Three examples of elementary block diagrams

Negative feedback

Figure 3.8(c) shows a more complicated case. Here the two blocks are connected in a feedback arrangement so that each feeds into the other. When the feedback Y2 (s) is subtracted, as shown in the figure, we call it negative feedback. As you will see, negative feedback is usually required for system stability. For now we will simply solve the equations and then relate them back to the diagram. The equations are U1 (s) = R(s) − Y2 (s), Y2 (s) = G2 (s)G1 (s)U1 (s), Y1 (s) = G1 (s)U1 (s), and their solution is

G1 (s) R(s). 1 + G1 (s)G2 (s) We can express the solution by the following rule: Y1 (s) =

(3.50)

The gain of a single-loop negative feedback system is given by the forward gain divided by the sum of 1 plus the loop gain. Positive feedback

Unity feedback system

When the feedback is added instead of subtracted, we call it positive feedback. In this case, the gain is given by the forward gain divided by the sum of 1 minus the loop gain. The three elementary cases given in Fig. 3.8 can be used in combination to solve, by repeated reduction, any transfer function defined by a block diagram. However, the manipulations can be tedious and subject to error when the topology of the diagram is complicated. Figure 3.9 shows examples of block-diagram algebra that complement those shown in Fig. 3.8. Figures 3.9(a) and (b) show how the interconnections of a block diagram can be manipulated without affecting the mathematical relationships. Figure 3.9(c) shows how the manipulations can be used to convert a general system (on the left) to a system without a component in the feedback path, usually referred to as a unity feedback system. In all cases the basic principle is to simplify the topology while maintaining exactly the same relationships among the remaining variables of the block diagram. In relation to the algebra of the underlying linear equations, block-diagram reduction is a pictorial way to solve equations by eliminating variables.

3.2 System Modeling Diagrams Figure 3.9 Examples of block-diagram algebra

G1

U1

Y1




G1

U1

105

Y1

Pick-off point

1 G1 Y2

Y2 (a)



U1




G1

Y1






G1

U1






Y1



U2

U2

G1

R

1 G2

(b)

R






U1

Y1 G1

Y

 Y2

G2









G2

G1

Y



U2 (c)

Figure 3.10 Block diagram of a second-order system

2 R








4 s








1 s

Y

1 s

Y

(a)

R

 




2s  4 s

(b)

EXAMPLE 3.20

Transfer Function from a Simple Block Diagram Find the transfer function of the system shown in Fig. 3.10(a). Solution. First we simplify the block diagram by reducing the parallel combination of the controller path. This results in the diagram of Fig. 3.10(b), and we use the feedback rule to obtain the closed-loop transfer function: 2s+4

T (s) =

Y (s) 2s + 4 s2 = . = 2 2s+4 R(s) s + 2s + 4 1 + s2

106 Chapter 3 Dynamic Response Transfer Function from the Block Diagram EXAMPLE 3.21 Find the transfer function of the system shown in Fig. 3.11(a). Solution. First we simplify the block diagram. Using the principles of Eq. (3.50), we replace the feedback loop involving G1 and G3 by its equivalent transfer function, noting that it is a positive feedback loop. The result is Fig. 3.11(b). The next step is to move the pick-off point preceding G2 to its output [see Fig. 3.11(a)], as shown in Fig. 3.11(c). The negative feedback loop on the left is in series with the subsystem on the right, which is composed of the two parallel blocks G5 and G6 /G2 . The overall transfer function can be written using all three rules for reduction given by Fig. 3.8:   G1 G2 G6 Y (s) 1−G1 G3 T (s) = = G5 + G1 G2 G4 R(s) G2 1 + 1−G G 1 3

G1 G2 G5 + G1 G6 . = 1 − G1 G3 + G1 G2 G4

Figure 3.11 Example for block-diagram simplification

G6 R













G1

G5

G2








Y

 G3 G4 (a)

G6 R








G1 1 G1G3

G2

G5






 Y

G4 (b) G6 G2 R

 




G1 1 G1G3

G2 G4 (c)

G5

 




Y

3.2 System Modeling Diagrams

107

As we have seen, a system of algebraic equations may be represented by a block diagram that represents individual transfer functions by blocks and has interconnections that correspond to the system equations. A block diagram is a convenient tool to visualize the system as a collection of interrelated subsystems that emphasize the relationships among the system variables.

3.2.2 Block Diagram Reduction Using MATLAB If the individual transfer functions are available for components in a control system, it is possible to use MATLAB commands to compute the transfer functions of interconnected systems. The three commands series, parallel, and feedback can be used for this purpose. They compute the transfer functions of two component block transfer functions in series, parallel, and feedback configurations, respectively. The next simple example illustrates their use.

EXAMPLE 3.22

Transfer Function of a Simple System Using MATLAB Repeat the computation of the transfer function for the block diagram in Fig. 3.10(a) using MATLAB. Solution. We label the transfer function of the separate blocks shown in Fig. 3.10(a) as illustrated in Fig. 3.12. Then we combine the two parallel blocks G1 and G2 by num1=[2]; % form G1 den1=[1]; sysG1=tf(num1,den1); % define subsystem G1 num2=[4]; % form G2 den2=[1 0]; sysG2=tf(num2,den2); % define subsystem G2 % parallel combination of G1 and G2 to form subsystem G3 sysG3=parallel(sysG1,sysG2); then we combine the result G3, with the G4 in series by num4=[1]; den4=[1 0]; sysG4=tf(num4,den4); sysG5=series(sysG3,sysG4);

% form G4 % define subsystem G4 % series combination of G3 and G4

and complete the reduction of the feedback system by num6=[1]; den6=[1]; Figure 3.12 Example for block-diagram simplification

% form G6

G1 R






G2

 G6








G4

Y

108 Chapter 3 Dynamic Response sysG6=tf(num6,den6) [sysCL]=feedback(sysG5,sysG6,-1)

% define subsystem G6 % feedback combination of G5 and G6

The MATLAB results are sysCL of the form 2s + 4 Y (s) = 2 R(s) s + 2s + 4 and this is the same result as the one obtained by block diagram reduction.

3.3 Effect of Pole Locations Once the transfer function has been determined by any of the available methods, we can start to analyze the response of the system it represents. When the system equations are simultaneous ordinary differential equations (ODEs), the transfer function that results will be a ratio of polynomials; that is, H(s) = b(s)/a(s).

Poles Zeros

The impulse response is the natural response.

First-order system impulse response

Stability

Time constant τ

If we assume that b and a have no common factors (as is usually the case), then values of s such that a(s) = 0 will represent points where H(s) is infinity. As we saw in Section 3.1.5, these s-values are called poles of H(s). Values of s such that b(s) = 0 are points where H(s) = 0 and the corresponding s-locations are called zeros. The effect of zeros on the transient response will be discussed in Section 3.5. These poles and zeros completely describe H(s) except for a constant multiplier. Because the impulse response is given by the time function corresponding to the transfer function, we call the impulse response the natural response of the system. We can use the poles and zeros to compute the corresponding time response and thus identify time histories with pole locations in the s-plane. For example, the poles identify the classes of signals contained in the impulse response, as may be seen by a partial-fraction expansion of H(s). For a first-order pole, 1 . H(s) = s+σ Table A.2, entry 7, indicates that the impulse response will be an exponential function; that is, h(t) = e−σ t 1(t). When σ > 0, the pole is located at s < 0, the exponential expression decays, and we say the impulse response is stable. If σ < 0, the pole is to the right of the origin. Because the exponential expression here grows with time, the impulse response is referred to as unstable (Section 3.6). Figure 3.13(a) shows a typical stable response and defines the time constant τ = 1/σ (3.51) as the time when the response is 1/e times the initial value. Hence, it is a measure of the rate of decay. The straight line is tangent to the exponential curve at t = 0 and terminates at t = τ . This characteristic of an exponential expression is useful in sketching a time plot or checking computer results. Figure 3.13(b) shows the impulse and step response for a first-order system computed using MATLAB.

3.3 Effect of Pole Locations Figure 3.13 First-order system response: (a) impulse response; (b) impulse response and step response using MATLAB


109

1.0 0.8 0.6

est

h(t)

1 e

0.4 0.2 0

0

1.0

2.0 Time (sec) (a) tt

3.0

4.0

3

4

1 y

h(t), y(t)

0.8 0.6 0.4 h

0.2 0

0

1

2 Time (sec) (b)

EXAMPLE 3.23

Response versus Pole Locations, Real Roots Compare the time response with the pole locations for the system with a transfer function between input and output given by 2s + 1 H(s) = 2 . (3.52) s + 3s + 2 Solution. The numerator is   1 b(s) = 2 s + , 2 and the denominator is a(s) = s2 + 3s + 2 = (s + 1)(s + 2). The poles of H(s) are therefore at s = −1 and s = −2 and the one (finite) zero is at s = − 21 . A complete description of this transfer function is shown by the plot of the locations of the poles and the zeros in the s-plane using the MATLAB pzmap(num,den) function with

110 Chapter 3 Dynamic Response Figure 3.14 Sketch of s-plane showing poles as crosses and zeros as circles

Im(s) j

2

1

 Zero  Pole

1

Re(s)

j

num=[2 1]; den=[1 3 2]; (see Fig. 3.14). A partial-fraction expansion of H(s) results in 3 1 + . s+1 s+2 From Table A.2 we can look up the inverse of each term in H(s), which will give us the time function h(t) that would result if the system input were an impulse. In this case,  −e−t + 3e−2t t ≥ 0, h(t) = (3.53) 0 t < 0. H(s) = −

“Fast poles” and “slow poles” refer to relative rate of signal decay.

Impulse response using MATLAB

We see that the shape of the component parts of h(t), which are e−t and e−2t , are determined by the poles at s = −1 and −2. This is true of more complicated cases as well: In general, the shapes of the components of the natural response are determined by the locations of the poles of the transfer function. A sketch of these pole locations and corresponding natural responses is given in Fig. 3.15, along with other pole locations including complex ones, which will be discussed shortly. The role of the numerator in the process of partial-fraction expansion is to influence the size of the coefficient that multiplies each component. Because e−2t decays faster than e−t , the signal corresponding to the pole at −2 decays faster than the signal corresponding to the pole at −1. For brevity we simply say that the pole at −2 is faster than the pole at −1. In general, poles farther to the left in the s-plane are associated with natural signals that decay faster than those associated with poles closer to the imaginary axis. If the poles had been located with positive values of s (in the right half of the s-plane), the response would have been a growing exponential function and thus unstable. Figure 3.16 shows that the fast 3e−2t term dominates the early part of the time history and that the −e−t term is the primary contributor later on. The purpose of this example is to illustrate the relationship between the poles and the character of the response, which can be done exactly only by finding the inverse Laplace transform and examining each term as before. However, if we simply wanted to plot the impulse response for this example, the expedient way would be to use the MATLAB sequence numH = [2 1]; denH = [1 3 2];

% form numerator % form denominator

3.3 Effect of Pole Locations Figure 3.15 Time functions associated with points in the s-plane (LHP, left half-plane; RHP, right half-plane)

111

Im(s) STABLE

UNSTABLE

LHP

RHP

Re(s)

Figure 3.16 Impulse response of Example 3.23 [Eq. (3.52)]

2.0 1.5 1.0 h(t) 0.5 0 0.5

0

1

2

3 4 Time (sec)

sysH=tf(numH,denH); impulse(sysH);

5

6

% define system from its numerator and denominator % compute impulse response

The result is shown in Fig. 3.16. Complex poles can be defined in terms of their real and imaginary parts, traditionally referred to as s = −σ ± jωd . This means that a pole has a negative real part if σ is positive. Since complex poles always come in complex conjugate pairs, the denominator corresponding to a complex pair will be a(s) = (s + σ − jωd )(s + σ + jωd ) = (s + σ )2 + ωd2 . (3.54) When finding the transfer function from differential equations, we typically write the result in the polynomial form ωn2 H(s) = 2 . (3.55) s + 2ζ ωn s + ωn2

112 Chapter 3 Dynamic Response By multiplying out the form given by Eq. (3.54) and comparing it with the coefficients of the denominator of H(s) in Eq. (3.55), we find the correspondence between the parameters to be  σ = ζ ωn and ωd = ωn 1 − ζ 2 , (3.56) Damping ratio; damped and undamped natural frequency

where the parameter ζ is the damping ratio7 and ωn is the undamped natural frequency. The poles of this transfer function are located at a radius ωn in the s-plane and at an angle θ = sin−1 ζ , as shown in Fig. 3.17. Therefore, the damping ratio reflects the level of damping as a fraction of the critical damping value where the poles become real. In rectangular coordinates the poles are at s = −σ ± jωd . When ζ = 0, we have no damping, θ = 0, and the damped natural frequency ωd = ωn , the undamped natural frequency. For purposes of finding the time response from Table A.2 corresponding to a complex transfer function, it is easiest to manipulate the H(s) so that the complex poles fit the form of Eq. (3.54), because then the time response can be found directly from the table. Equation (3.55) can be rewritten as H(s) =

Standard second-order system impulse response

(s + ζ ωn

ωn2 . + ωn2 (1 − ζ 2 )

(3.57)

Therefore, from entry number 20 in Table A.2 and the definitions in Eq. (3.56), we see that the impulse response is h(t) = 

Figure 3.17 s-plane plot for a pair of complex poles

)2

ωn 1 − ζ2

e−σ t (sin ωd t)1(t).

(3.58)

Im(s)

u  sin1z

vn Re(s) s

vd

7 In communications and filter engineering, the standard second-order transfer function is written as H = 1/[1 + Q(s/ωn + ωn /s)]. Here, ωn is called the band center and Q is the quality factor. Comparison with Eq. (3.55) shows that Q = 1/2ζ .

3.3 Effect of Pole Locations

113

Figure 3.18(a) plots h(t) for several values of ζ such that time has been normalized to the undamped natural frequency ωn . Note that the actual frequency ωd decreases slightly as the damping ratio increases. Note also that for very low damping the response is oscillatory, while for large damping (ζ near 1) the response shows no oscillation. A few of these responses are sketched in Fig. 3.15 to show qualitatively how changing pole locations in the s-plane affect impulse responses. You will find it useful as a control designer to commit the image of Fig. 3.15 to memory so that you can understand instantly how changes in pole locations influence the time response. Figure 3.18 Responses of second-order systems versus ζ : (a) impulse responses; (b) step responses

z0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

1.0 0.8 0.6 0.4 0.2 0.9 y(t)

0.0

z1

0.2 0.4 0.6 0.8 1.0

0

2

4

6 vnt

8

10

12

8

10

12

(a) 2.0

z0 0.1 0.2 0.3 0.4 0.5 0.6

1.8 1.6 1.4 1.2 y(t)

1.0 0.8 z  0.7 0.8 0.9 1.0

0.6 0.4 0.2 0

0

2

4

6 vnt (b)

114 Chapter 3 Dynamic Response Stability depends on whether natural response grows or decays.

Step response

Three pole locations are shown in Fig. 3.19 for comparison with the corresponding impulse responses in Fig. 3.18(a). The negative real part of the pole, σ , determines the decay rate of an exponential envelope that multiplies the sinusoid, as shown in Fig. 3.20. Note that if σ < 0 (and the pole is in the RHP), then the natural response will grow with time, so, as defined earlier, the system is said to be unstable. If σ = 0, the natural response neither grows nor decays, so stability is open to debate. If σ > 0, the natural response decays, so the system is stable. It is also interesting to examine the step response of H(s)—that is, the response of the system H(s) to a unit step input u = 1(t), where U(s) = 1/s. The step-response transform is given by Y (s) = H(s)U(s), which is found in Table A.2, entry 21. Figure 3.18(b), which plots y(t) for several values of ζ , shows that the basic transient response characteristics from the impulse response carry over quite well to the step response; the difference between the two responses is that the step response’s final value is the commanded unit step.

Figure 3.19 Pole locations corresponding to three values of ζ

Im(s)

Im(s)

45

30

17.5

Re(s)

Re(s)

z  0.707

Figure 3.20 Second-order system response with an exponential envelope

Im(s)

Re(s)

z  0.5

z  0.3

1 0.8 0.6

est

0.4 h(t)

0.2 0 0.2

est

0.4 0.6 0.8 1

0

5

10

15 Time (sec)

20

25

30

3.3 Effect of Pole Locations

EXAMPLE 3.24

115

Oscillatory Time Response Discuss the correlation between the poles of H(s) =

s2

2s + 1 + 2s + 5

(3.59)

and the impulse response of the system and find the exact impulse response. Solution. From the form of H(s) given by Eq. (3.55), we see that √ ωn2 = 5 ⇒ ωn = 5 = 2.24 rad/sec and 1 2ζ ωn = 2 ⇒ ζ = √ = 0.447. 5 This indicates that we should expect a frequency of around 2 rad/sec with very little oscillatory motion. In order to obtain the exact response, we manipulate H(s) until the denominator is in the form of Eq. (3.54): H(s) =

2s + 1 2s + 1 = . s2 + 2s + 5 (s + 1)2 + 22

From this equation we see that the poles of the transfer function are complex, with real part −1 and imaginary parts ±2j. Table A.2 has two entries, numbers 19 and 20, that match the denominator. The right side of the preceding equation needs to be broken into two parts so that they match the numerators of the entries in the table: H(s) =

2s + 1 2 s+1 1 =2 − . 2 2 2 2 (s + 1) + 2 (s + 1) + 2 2 (s + 1)2 + 22

Thus, the impulse response is   1 h(t) = 2e−t cos 2t − e−t sin 2t 1(t). 2

Impulse response by MATLAB

Figure 3.21 is a plot of the response and shows how the envelope attenuates the sinusoid, the domination of the 2 cos 2t term, and the small phase shift caused by the − 21 sin 2t term. As in the previous example, the expedient way of determining the impulse response would be to use the MATLAB sequence numH = [2 1]; denH = [1 2 5]; sysH=tf(numH,denH); t=0:0.1:6; y=impulse(sysH,t); plot(t,y); as shown in Fig. 3.21.

% form numerator % form denominator % define system by its numerator and denominator % form time vector % compute impulse response % plot impulse response

116 Chapter 3 Dynamic Response Figure 3.21 System response for Example 3.24

2 1.5 1

h(t)

0.5 0 0.5 1 1.5 2

0

1

2

3 Time (sec)

4

5

6

3.4 Time-Domain Specifications Definitions of rise time, settling time, overshoot, and peak time

Specifications for a control system design often involve certain requirements associated with the time response of the system. The requirements for a step response are expressed in terms of the standard quantities illustrated in Fig. 3.22: 1. The rise time tr is the time it takes the system to reach the vicinity of its new set point. 2. The settling time ts is the time it takes the system transients to decay. 3. The overshoot Mp is the maximum amount the system overshoots its final value divided by its final value (and is often expressed as a percentage). 4. The peak time tp is the time it takes the system to reach the maximum overshoot point.

3.4.1 Rise Time

Rise time tr

For a second-order system, the time responses shown in Fig. 3.18(b) yield information about the specifications that is too complex to be remembered unless converted to a simpler form. By examining these curves in light of the definitions given in Fig. 3.22, we can relate the curves to the pole-location parameters ζ and ωn . For example, all the curves rise in roughly the same time. If we consider the curve for ζ = 0.5 to be an average, the rise time from y = 0.1 to y = 0.9 is approximately ωn tr = 1.8. Thus we can say that 1.8 . (3.60) tr ∼ = ωn Although this relationship could be embellished by including the effect of the damping ratio, it is important to keep in mind how Eq. (3.60) is typically used. It is accurate only for a second-order system with no zeros; for all other systems it is a rough approximation to the relationship between tr and ωn . Most systems being analyzed for control systems design are more complicated than the pure second-order system, so designers use Eq. (3.60) with the knowledge that it is a rough approximation only.

3.4 Time-Domain Specifications Figure 3.22 Definition of rise time tr , settling time ts , and overshoot Mp

tp

117

Mp 1%

1 0.9

0.1 t

tr ts

3.4.2 Overshoot and Peak Time For the overshoot Mp we can be more analytical. This value occurs when the derivative is zero, which can be found from calculus. The time history of the curves in Fig. 3.18(b), found from the inverse Laplace transform of H(s)/s, is   σ −σ t cos ωd t + sin ωd t , (3.61) y(t) = 1 − e ωd  where ωd = ωn 1 − ζ 2 and σ = ζ ωn . We may rewrite the preceding equation using the trigonometric identity A sin(α) + B cos(α) = C cos(α − β) or C=



A2 + B 2 = 

β = tan Standard second-order system step response

with A =

σ ωd ,

−1

1 1 − ζ2 

  A = tan−1 B

, 

ζ 1 − ζ2

 ,

B = 1, and α = ωd t, in a more compact form as y(t) = 1 − 

e−σ t

cos(ωd t − β). (3.62) 1 − ζ2 When y(t) reaches its maximum value, its derivative will be zero:   σ y˙ (t) = σ e−σ t cos ωd t + sin ωd t − e−σ t (−ωd sin ωd t + σ cos ωd t) = 0 ωd  2  σ + ωd sin ωd t = 0. = e−σ t ωd This occurs when sin ωd t = 0, so ωd tp = π and thus tp =

π . ωd

(3.63)

118 Chapter 3 Dynamic Response Peak time tp

Substituting Eq. (3.63) into the expression for y(t), we compute   σ −σ π/ωd cos π + y(tp ) = 1 + Mp = 1 − e sin π ωd = 1 + e−σ π/ωd .

Overshoot Mp

Thus we have the formula

√ 2 Mp = e−π ζ / 1−ζ ,

0 ≤ ζ < 1,

(3.64)

which is plotted in Fig. 3.23. Two frequently used values from this curve are Mp = 0.16 for ζ = 0.5 and Mp = 0.05 for ζ = 0.7.

3.4.3 Settling Time

Settling time ts

The final parameter of interest from the transient response is the settling time ts . This is the time required for the transient to decay to a small value so that y(t) is almost in the steady state. Various measures of smallness are possible. For illustration we will use 1% as a reasonable measure; in other cases 2% or 5% are used. As an analytic computation, we notice that the deviation of y from 1 is the product of the decaying exponential e−σ t and the circular functions sine and cosine. The duration of this error is essentially decided by the transient exponential, so we can define the settling time as that value of ts when the decaying exponential reaches 1%: e−ζ ωn ts = 0.01. Therefore, ζ ωn ts = 4.6, or

4.6 4.6 , (3.65) = ζ ωn σ where σ is the negative real part of the pole, as may be seen from Fig. 3.17. ts =

Figure 3.23 Overshoot Mp versus damping ratio ζ for the second-order system

100 90 80 70 Mp, %

60 50 40 30

16%

20 10 0 0.0

5% 0.2

0.4

0.6 z

0.8

1.0

3.4 Time-Domain Specifications

119

Equations (3.60), (3.64), and (3.65) characterize the transient response of a system having no finite zeros and two complex poles and with undamped natural frequency ωn , damping ratio ζ , and negative real part σ . In analysis and design, they are used to estimate rise time, overshoot, and settling time, respectively, for just about any system. In design synthesis we wish to specify tr , Mp , and ts and to ask where the poles need to be so that the actual responses are less than or equal to these specifications. For specified values of tr , Mp , and ts , the synthesis form of the equation is then 1.8 , (3.66) ωn ≥ tr ζ ≥ ζ (Mp ) (from Fig. 3.23), (3.67)

Design synthesis

4.6 . (3.68) ts These equations, which can be graphed in the s-plane as shown in Fig. 3.24(a–c), will be used in later chapters to guide the selection of pole and zero locations to meet control system specifications for dynamic response. It is important to keep in mind that Eqs. (3.66)–(3.68) are qualitative guides and not precise design formulas. They are meant to provide only a starting point for the design iteration. After the control design is complete, the time response should always be checked by an exact calculation, usually by numerical simulation, to verify whether the time specifications have actually been met. If not, another iteration of the design is required. For a first-order system, σ , H(s) = s+σ and the transform of the step response is σ Y (s) = . s(s + σ ) We see from entry 11 in Table A.2 that Y (s) corresponds to σ ≥

First-order system step response

y(t) = (1 − e−σ t )1(t).

Im(s)

sin1z

Im(s)

(3.69)

Im(s)

Im(s)

s vn Re(s)

(a)

Re(s)

(b)

Re(s)

(c)

Re(s)

(d)

Figure 3.24 Graphs of regions in the s-plane delineated by certain transient requirements: (a) rise time; (b) overshoot; (c) settling time; (d) composite of all three requirements

120 Chapter 3 Dynamic Response

Time constant τ

EXAMPLE 3.25

Comparison with the development for Eq. (3.65) shows that the value of ts for a first-order system is the same: 4.6 . ts = σ No overshoot is possible, so Mp = 0. The rise time from y = 0.1 to y = 0.9 can be seen from Fig. 3.13 to be 2.2 ln 0.9 − ln 0.1 tr = = . σ σ However, it is more typical to describe a first-order system in terms of its time constant, which was defined in Fig. 3.13 to be τ = 1/σ .

Transformation of the Specifications to the s-Plane Find the allowable regions in the s-plane for the poles of a transfer function of a system if the system response requirements are tr ≤ 0.6 sec, Mp ≤ 10%, and ts ≤ 3 sec. Solution. Without knowing whether or not the system is second order with no zeros, it is impossible to find the allowable region accurately. Regardless of the system, we can obtain a first approximation using the relationships for a second-order system. Equation (3.66) indicates that 1.8 ωn ≥ = 3.0 rad/sec, tr Eq. (3.67) and Fig. 3.23 indicate that ζ ≥ 0.6, and Eq. (3.68) indicates that 4.6 σ ≥ = 1.5 sec. 3 The allowable region is anywhere to the left of the solid line in Fig. 3.25. Note that any pole meeting the ζ and ωn restrictions will automatically meet the σ restriction.

3.5 Effects of Zeros and Additional Poles

Effect of zeros The effect of zeros near poles

Relationships such as those shown in Fig. 3.24 are correct for the simple second-order system; for more complicated systems they can be used only as guidelines. If a certain design has an inadequate rise time (is too slow), we must raise the natural frequency; if the transient has too much overshoot, then the damping needs to be increased; if the transient persists too long, the poles need to be moved to the left in the s-plane. Thus far only the poles of H(s) have entered into the discussion. There may also be zeros of H(s).8 At the level of transient analysis, the zeros exert their influence by modifying the coefficients of the exponential terms whose shape is decided by the poles, as seen in Example 3.23. To illustrate this further, consider the following two transfer functions, which have the same poles but different zeros: 8 We assume that b(s) and a(s) have no common factors. If this is not so, it is possible for b(s) and a(s)

to be zero at the same location and for H(s) to not equal zero there. The implications of this case will be discussed in Chapter 7, when we have a state-space description.

3.5 Effects of Zeros and Additional Poles Figure 3.25 Allowable region in s-plane for Example 3.25

121

3 2 1 Im(s) 0 1 2 3 5

4

3

2 1 Re(s)

0

1

2 (s + 1)(s + 2) 2 2 − , = s+1 s+2 2(s + 1.1) H2 (s) = 1.1(s + 1)(s + 2)   0.1 0.9 2 = + 1.1 s + 1 s + 2 H1 (s) =

=

0.18 1.64 + . s+1 s+2

(3.70)

(3.71)

They are normalized to have the same DC gain (i.e., gain at s = 0). Notice that the coefficient of the (s + 1) term has been modified from 2 in H1 (s) to 0.18 in H2 (s). This dramatic reduction is brought about by the zero at s = −1.1 in H2 (s), which almost cancels the pole at s = −1. If we put the zero exactly at s = −1, this term will vanish completely. In general, a zero near a pole reduces the amount of that term in the total response. From the equation for the coefficients in a partial-fraction expansion, Eq. (3.43), C1 = (s − p1 )F(s)|s=p1 , we can see that if F(s) has a zero near the pole at s = p1 , the value of F(s) will be small because the value of s is near the zero. Therefore, the coefficient C1 , which reflects how much of that term appears in the response, will be small. In order to take into account how zeros affect the transient response when designing a control system, we consider transfer functions with two complex poles and one zero. To expedite the plotting for a wide range of cases, we write the transform in a form with normalized time and zero locations: H(s) =

(s/αζ ωn ) + 1 . (s/ωn )2 + 2ζ (s/ωn ) + 1

(3.72)

122 Chapter 3 Dynamic Response The zero is located at s = −αζ ωn = −ασ . If α is large, the zero will be far removed from the poles and the zero will have little effect on the response. If α ∼ = 1, the value of the zero will be close to that of the real part of the poles and can be expected to have a substantial influence on the response. The step-response curves for ζ = 0.5 and for several values of α are plotted in Fig. 3.26. We see that the major effect of the zero is to increase the overshoot Mp , whereas it has very little influence on the settling time. A plot of Mp versus α is given in Fig. 3.27. The plot shows that the zero has very little effect on Mp if α > 3, but as α decreases below 3, it has an increasing effect, especially when α = 1 or less.

1.8 1.6 1.4 Step response of H(s)

Figure 3.26 Plots of the step response of a second-order system with a zero (ζ = 0.5)

1.2 1.0 a 1 2 4 100

0.8 0.6 0.4 0.2 0

0

2

4

6

8

10

vnt

Figure 3.27 Plot of overshoot Mp as a function of normalized zero location α. At α = 1, the real part of the zero equals the real part of the poles

2.0 1.8 1.6 1.4 1.2 z0.3 0.5 0.7

Mp 1.0 0.8 0.6 0.4 0.2 0

0

2

4

6 a

8

10

3.5 Effects of Zeros and Additional Poles Figure 3.28 Second-order step responses y(t) of the transfer functions H(s), H0 (s), and Hd (s)

123

1.8 1.6

y(t)

1.4 1.2 1.0 y (t)

y0 (t)

0.8 0.6

yd (t)

0.4 0.2 0 ⫺0.2

0

2

4

6

8

10

Time (sec)

RHP or nonminimum-phase zero

EXAMPLE 3.26

Figure 3.26 can be explained in terms of Laplace-transform analysis. First we replace s/ωn with s: s/αζ + 1 H(s) = 2 . s + 2ζ s + 1 This has the effect of normalizing frequency in the transfer function and normalizing time in the corresponding step responses; thus τ = ωn t. We then rewrite the transfer function as the sum of two terms: 1 1 s H(s) = 2 + . (3.73) 2 s + 2ζ s + 1 αζ s + 2ζ s + 1 The first term, which we shall call H0 (s), is the original term (having no finite zero), and the second term Hd (s), which is introduced by the zero, is a product of a constant (1/αζ ) times s times the original term. The Laplace transform of df /dt is sF(s), so Hd (s) corresponds to a product of a constant times the derivative of the original term, i.e., 1 . y (t). y(t) = y0 (t) + yd (t) = y0 (t) + αζ 0 The step responses of H0 (s) denoted by y0 (t) and Hd (s) denoted by yd (t) are plotted in Fig. 3.28. Looking at these curves, we can see why the zero increased the overshoot: The derivative has a large hump in the early part of the curve, and adding this to the H0 (s) response lifts up the total response of H(s) to produce the overshoot. This analysis is also very informative for the case when α < 0 and the zero is in the RHP where s > 0. (This is typically called an RHP zero and is sometimes referred to as a nonminimum-phase zero, a topic to be discussed in more detail in Section 6.1.1.) In this case the derivative term is subtracted rather than added. A typical case is sketched in Fig. 3.29.

Effect of the Proximity of the Zero to the Pole Locations on the Transient Response Consider the second-order system with a finite zero and unity DC gain, (s + z) 24 . H(s) = z (s + 4)(s + 6)

124 Chapter 3 Dynamic Response Figure 3.29 Step responses y(t) of a second-order system with a zero in the RHP: a nonminimum-phase system

1.5 1.0 y(t) 0.5 y0 (t) 0

y (t)

0.5 yd (t) 1.0 1.5

0

2

4

6

8

10

Time (sec)

Determine the effect of the zero location (s = −z) on the unit step response when z = {1, 2, 3, 4, 5, 6}. Solution. The step response is the inverse Laplace transform of H1 (s) = H(s)

24 (s + z) 24 s 24 1 = = + s z s(s + 4)(s + 6) z s(s + 4)(s + 6) s(s + 4)(s + 6)

and is the sum of the two parts, y(t) = y1 (t) + y2 (t), where 12 −4t 12 −6t e − e , z z

t y1 (τ )dτ = −3e−4t + 2e−6t + 1, y2 (t) = z y1 (t) =

0

and

 y(t) = 1 +

   12 12 −6t − 3 e−4t + 2 − e . z z

It is seen that if z = 4 or z = 6, one of the modes of the system is absent from the output, and the response is first order due to the pole–zero cancellations. The step responses of the system is shown in Fig. 3.30 (z = 4, dashed, z = 6 dot dashed). It is seen that the effect of the zero is most pronounced in terms of the additional overshoot for z = 1 (zero location closest to the origin). The system also has overshoot for z = 2, 3. For z = 4 or z = 6 the responses are first order as expected. It is interesting that for z = 5, where the zero is located between the two poles, there is no overshoot.

3.5 Effects of Zeros and Additional Poles Figure 3.30 Effect of zero on transient response

125

2.5 z=1

Unit step response

2

1.5 z=2 z=3

1

z=6

0.5

0

0

0.5

1

1.5

2

2.5

Time (sec)

EXAMPLE 3.27

Effect of the Proximity of the Complex Zeros to the Lightly Damped Poles Consider the third-order feedback system with a pair of lightly damped poles and a pair of complex zeros with the transfer function, H(s) =

(s + α)2 + β 2  . (s + 1) (s + 0.1)2 + 1

Determine the effect of the complex zero locations (s = −α ± jβ) on the unit step response of the system for the three different zero locations (α, β) = (0.1, 1.0), (α, β) = (0.25, 1.0), and (α, β) = (0.5, 1.0) as shown in Fig. 3.31. Solution. We plot the three unit step responses using MATLAB as shown in Fig. 3.32. The effect of the lightly damped modes are clearly seen as oscillations in the step responses for the cases where (α, β) = (0.25, 1.0) or (α, β) = (0.5, 1.0), that is, when the complex zeros are not close to the locations of the lightly damped poles as shown in Fig. 3.31. On the other hand, if the complex zeros cancel the lightly damped poles exactly as is the case for (α, β) = (0.1, 1.0), the oscillations are completely eliminated in the step response. In practice, the locations of the lightly damped poles are not known precisely and exact cancellation is not really possible. However, placing the complex zeros near the locations of the lightly damped poles may provide sufficient improvement in step response performance. We will come back to this technique later in Chapters 5, 7, and 10 in the context of dynamic compensator design.

126 Chapter 3 Dynamic Response Figure 3.31 Locations of complex zeros

jω +1j

−1.0

σ

−0.5 −0.25 0


1j

Figure 3.32 Effect of complex zeros on transient response

1.4

α = 0.5

1.2

α = 0.25

Unit step response

1

0.8

α = 0.1

0.6

0.4

0.2

0

0

5

10 Time (sec)

15

20

3.5 Effects of Zeros and Additional Poles

EXAMPLE 3.28

127

Aircraft Response Using MATLAB The transfer function between the elevator and altitude of the Boeing 747 aircraft described in Section 10.3.2 can be approximated as 30(s − 6) h(s) = . 2 δe (s) s(s + 4s + 13) 1. Use MATLAB to plot the altitude time history for a 1◦ impulsive elevator input. Explain the response, noting the physical reasons for the nonminimum-phase nature of the response. 2. Examine the accuracy of the approximations for tr , ts , and Mp [Eqs. (3.60) and (3.65) and Fig. 3.23]. Solution 1. The MATLAB statements to create the impulse response for this case are u = −1; numG = u∗30∗[1 −6]; denG = [1 4 13 0]; sysG=tf(numG,denG) y=impulse(sysG); plot(y);

% u = delta e % form numerator % form denominator % define system by its numerator and denominator % compute impulse response; y = h % plot impulse response

The result is the plot shown in Fig. 3.33. Notice how the altitude drops initially and then rises to a new final value. The final value is predicted by the Final Value Theorem:  30(−6)(−1) 30(s − 6)(−1)  = +13.8. = h(∞) = s  2 s(s + 4s + 13) s=0 13 Response of a nonminimum-phase system

The fact that the response has a finite final value for an impulsive input is due to the s-term in the denominator. This represents a pure integration and the integral of an impulse function is a finite value. If the input had been a step, the altitude

Figure 3.33 Response of an airplane’s altitude to an impulsive elevator input

16 14 12 Altitude (ft)

10 8 6 4 2 0
2

0

1

2

3 Time (sec)

4

5

6

128 Chapter 3 Dynamic Response would have continued to increase with time; in other words the integral of a step function is a ramp function. The initial drop is predicted by the RHP zero in the transfer function. The negative elevator deflection is defined to be upward by convention (see Fig. 10.30). The upward deflection of the elevators drives the tail down, which rotates the craft nose up and produces the climb. The deflection at the initial instant causes a downward force before the craft has rotated; therefore, the initial altitude response is down. After rotation, the increased lift resulting from the increased angle of attack of the wings causes the airplane to climb. 2. The rise time from Eq. (3.60) is 1.8 1.8 = √ = 0.5 sec. tr = ωn 13 We find the damping ratio ζ from the relation 2 2ζ ωn = 4 ⇒ ζ = √ = 0.55. 13 From Fig. 3.23 we find the overshoot Mp to be 0.14. Because 2ζ ωn = 2σ = 4, [Eq. (3.65)] shows that 4.6 4.6 = = 2.3 sec. ts = σ 2 Detailed examination of the time history h(t) from MATLAB output shows that tr ∼ = 0.43 sec, Mp ∼ = 0.14, and ts ∼ = 2.6 sec, which are reasonably close to the estimates. The only significant effect of the nonminimum-phase zero was to cause the initial response to go in the “wrong direction” and make the response somewhat sluggish. Effect of extra pole

In addition to studying the effects of zeros, it is useful to consider the effects of an extra pole on the standard second-order step response. In this case, we take the transfer function to be 1 . (3.74) H(s) = (s/αζ ωn + 1)[(s/ωn )2 + 2ζ (s/ωn ) + 1] Plots of the step response for this case are shown in Fig. 3.34 for ζ = 0.5 and several values of α. In this case the major effect is to increase the rise time. A plot of the rise time versus α is shown in Fig. 3.35 for several values of ζ . From this discussion we can draw several conclusions about the dynamic response of a simple system as revealed by its pole–zero patterns:

Effects of Pole–Zero Patterns on Dynamic Response 1. For a second-order system with no finite zeros, the transient response parameters are approximated as follows: 1.8 , Rise time: tr ∼ = ωn ⎧ ⎨ 5%, ζ = 0.7, 16%, ζ = 0.5 (see Fig. 3.23), Overshoot: Mp ∼ = ⎩ 35%, ζ = 0.3,

3.5 Effects of Zeros and Additional Poles Figure 3.34 Step responses for several third-order systems with ζ = 0.5

129

1.2

1.0

0.8

y(t)

a1

0.6

2 5

0.4

100 0.2

0

2

0

4

6

8

10

vnt

Figure 3.35 Normalized rise time for several locations of an additional pole

9 8 7 6 vntr 5

z  1.0 0.7 0.5

4 3 2 1

0

1

2

3

4

5

6

7

8

9

10

a

4.6 . ts ∼ = σ 2. A zero in the left half-plane (LHP) will increase the overshoot if the zero is within a factor of 4 of the real part of the complex poles. A plot is given in Fig. 3.27. 3. A zero in the RHP will depress the overshoot (and may cause the step response to start out in the wrong direction). 4. An additional pole in the LHP will increase the rise time significantly if the extra pole is within a factor of 4 of the real part of the complex poles. A plot is given in Fig. 3.35. Settling time:

130 Chapter 3 Dynamic Response

3.6 Stability For nonlinear and time-varying systems, the study of stability is a complex and often difficult subject. In this section, we will consider only LTI systems for which we have the following condition for stability: An LTI system is said to be stable if all the roots of the transfer function denominator polynomial have negative real parts (i.e., they are all in the left hand s-plane) and is unstable otherwise.

Stable system

A system is stable if its initial conditions decay to zero and is unstable if they diverge. As just stated, an LTI (constant parameter) system is stable if all the poles of the system are strictly inside the left half s-plane [i.e., all the poles have negative real parts (s = −σ + jω, σ < 0)]. If any pole of the system is in the right half s-plane (i.e., has a positive real part, s = −σ + jω, σ > 0), then the system is unstable, as shown in Fig. 3.15. With any simple pole on the jω axis (σ = 0), small initial conditions will persist. For any other pole with σ = 0, oscillatory motion will persist. Therefore, a system is stable if its transient response decays and unstable if it does not. Figure 3.15 shows the time response of a system due to its pole locations. In later chapters we will address more advanced notions of stability, such as Nyquist’s frequency-response stability test (Chapter 6) and Lyapunov stability (Chapter 9).

3.6.1 Bounded Input–Bounded Output Stability A system is said to have bounded input–bounded output (BIBO) stability if every bounded input results in a bounded output (regardless of what goes on inside the system). A test for this property is readily available when the system response is given by convolution. If the system has input u(t), output y(t), and impulse response h(t), then

y(t) =

∞

−∞

h(τ )u(t − τ )dτ .

(3.75)

If u(t) is bounded, then there is a constant M such that |u| ≤ M < ∞, and the output is bounded by  

   |y| =  hu dτ 

≤ |h||u| dτ

∞ ≤M |h(τ )| dτ . −∞ ∞ Thus the output will be bounded if −∞ |h| dτ is bounded. On the other hand, suppose the integral is not bounded and the bounded input u(t − τ ) = +1 if h(τ ) > 0 and u(t − τ ) = −1 if h(τ ) < 0. In this case,

∞ y(t) = |h(τ )|dτ , (3.76) −∞

and the output is not bounded. We conclude that

3.6 Stability Figure 3.36 Capacitor driven by current source

Mathematical definition of BIBO stability

u(t)

C

131

+ y(t) −

The system with impulse response h(t) is BIBO-stable if and only if the integral

∞ |h(τ )|dτ < ∞. −∞

EXAMPLE 3.29

BIBO Stability for a Capacitor As an example, determine the capacitor driven by a current source sketched in Fig. 3.36. The capacitor voltage is the output and the current is the input. Solution. The impulse response of this setup is h(t) = 1(t), the unit step. Now for this response,

∞

∞ |h(τ )|dτ = dτ (3.77) −∞

Determination of BIBO stability by pole location

0

is not bounded. The capacitor is not BIBO-stable. Notice that the transfer function of the system is 1/s and has a pole on the imaginary axis. Physically we can see that constant input current will cause the voltage to grow, and thus the system response is neither bounded nor stable. In general, if an LTI system has any pole on the imaginary axis or in the RHP, the response will not be BIBO-stable; if every pole is inside the LHP, then the response will be BIBO-stable. Thus for these systems, pole locations of the transfer function can be used to check for stability. An alternative to computing the integral of the impulse response or even to locating the roots of the characteristic equation is given by Routh’s stability criterion, which we will discuss in Section 3.6.3.

3.6.2 Stability of LTI Systems Consider the LTI system whose transfer function denominator polynomial leads to the characteristic equation sn + a1 sn−1 + a2 sn−2 + · · · + an = 0. (3.78) Assume that the roots {pi } of the characteristic equation are real or complex, but are distinct. Note that Eq. (3.78) shows up as the denominator in the transfer function for the system as follows before any cancellation of poles by zeros is made: b0 sm + b1 sm−1 + · · · + bm Y (s) = T (s) = R(s) sn + a1 sn−1 + · · · + an m (s − zi ) K , m ≤ n. (3.79) = n i=1 i=1 (s − pi )

132 Chapter 3 Dynamic Response The solution to the differential equation whose characteristic equation is given by Eq. (3.78) may be written using partial-fraction expansion as

y(t) =

n 

Ki epi t ,

(3.80)

i=1

where {pi } are the roots of Eq. (3.78) and {Ki } depend on the initial conditions and zero locations. If a zero were to cancel a pole in the RHP for the transfer function, the corresponding Ki would equal zero in the output, but the unstable transient would appear in some internal variable. The system is stable if and only if (necessary and sufficient condition) every term in Eq. (3.80) goes to zero as t → ∞: epi t → 0

for all pi .

This will happen if all the poles of the system are strictly in the LHP, where Re{pi } < 0.

Internal stability occurs when all poles are strictly in the LHP The jω axis is the stability boundary

(3.81)

If any poles are repeated, the response must be changed from that of Eq. (3.80) by including a polynomial in t in place of Ki , but the conclusion is the same. This is called internal stability. Therefore, the stability of a system can be determined by computing the location of the roots of the characteristic equation and determining whether they are all in the LHP. If the system has any poles in the RHP, it is unstable. Hence the jω axis is the stability boundary between asymptotically stable and unstable response. If the system has nonrepeated jω axis poles, then it is said to be neutrally stable. For example, a pole at the origin (an integrator) results in a nondecaying transient. A pair of complex jω axis poles results in an oscillating response (with constant amplitude). If the system has repeated poles on the jω axis, then it is unstable [as it results in te±jωi t terms in Eq. (3.80)]. For example, a pair of poles at the origin (double integrator) results in an unbounded response. MATLAB software makes the computation of the poles, and therefore determination of the stability of the system, relatively easy. An alternative to locating the roots of the characteristic equation is given by Routh’s stability criterion, which we will discuss next.

3.6.3 Routh’s Stability Criterion There are several methods of obtaining information about the locations of the roots of a polynomial without actually solving for the roots. These methods were developed in the 19th century and were especially useful before the availability of MATLAB software. They are still useful for determining the ranges of coefficients of polynomials

133

3.6 Stability

for stability, especially when the coefficients are in symbolic (nonnumerical) form. Consider the characteristic equation of an nth-order system: 9 a(s) = sn + a1 sn−1 + a2 sn−2 + · · · + an−1 s + an .

A necessary condition for Routh stability

(3.82)

It is possible to make certain statements about the stability of the system without actually solving for the roots of the polynomial. This is a classical problem and several methods exist for the solution. A necessary condition for stability of the system is that all of the roots of Eq. (3.82) have negative real parts, which in turn requires that all the {ai } be positive.10 A necessary (but not sufficient) condition for stability is that all the coefficients of the characteristic polynomial be positive.

A necessary and sufficient condition for stability

If any of the coefficients are missing (are zero) or are negative, then the system will have poles located outside the LHP. This condition can be checked by inspection. Once the elementary necessary conditions have been satisfied, we need a more powerful test. Equivalent tests were independently proposed by Routh in 1874 and Hurwitz in 1895; we will discuss the former version. Routh’s formulation requires the computation of a triangular array that is a function of the {ai }. He showed that a necessary and sufficient condition for stability is that all of the elements in the first column of this array be positive. A system is stable if and only if all the elements in the first column of the Routh array are positive.

Routh array

To determine the Routh array, we first arrange the coefficients of the characteristic polynomial in two rows, beginning with the first and second coefficients and followed by the even-numbered and odd-numbered coefficients: sn : sn−1 :

1 a1

a2 a3

a4 a5

··· ...

We then add subsequent rows to complete the Routh array: Row Row Row Row

n n−1 n−2 n−3 .. .

sn : sn−1 : sn−2 : sn−3 : .. .

1 a1 b1 c1 .. .

a2 a3 b2 c2 .. .

Row Row Row

2 1 0

s2 : s: s0 :

* * *

*

a4 a5 b3 c3 .. .

··· ··· ··· ···

9 Without loss of generality, we can assume the polynomial to be monic (that is, the coefficient of the highest power of s is 1). 10 This is easy to see if we construct the polynomial as a product of first- and second-order factors.

134 Chapter 3 Dynamic Response We compute the elements from the (n − 2)th and (n − 3)th rows as follows:   1 a2 det a1 a3 a1 a2 − a3 b1 = − = , a1 a1   1 a4 det a1 a5 a1 a4 − a5 = , b2 = − a1 a1   1 a6 det a1 a7 a1 a6 − a7 = , b3 = − a1 a1   a1 a3 det b1 b2 b1 a3 − a1 b2 = , c1 = − b1 b1   a1 a5 det b1 b3 b1 a5 − a1 b3 = , c2 = − b1 b1   a1 a 7 det b1 b4 b1 a7 − a1 b4 c3 = − = . b1 b1 Note that the elements of the (n − 2)th row and the rows beneath it are formed from the two previous rows using determinants, with the two elements in the first column and other elements from successive columns. Normally there are n + 1 elements in the first column when the array terminates. If these are all positive, then all the roots of the characteristic polynomial are in the LHP. However, if the elements of the first column are not all positive, then the number of roots in the RHP equals the number of sign changes in the column. A pattern of +, −, + is counted as two sign changes: one change from + to − and another from − to +. For a simple proof of the Routh test, the reader is referred to Ho et al. (1998).

EXAMPLE 3.30

Routh’s Test The polynomial a(s) = s6 + 4s5 + 3s4 + 2s3 + s2 + 4s + 4 satisfies the necessary condition for stability since all the {ai } are positive and nonzero. Determine whether any of the roots of the polynomial are in the RHP. Solution. The Routh array for this polynomial is s6 :

1

3

1

4

s5 :

4 4·3−1·2 5 = 2 4

2

4

0

4·1−4·1 0= 4

4·4−1·0 4= 4

s4 :

3.6 Stability 5 ·4−4·4 12 = 2 − 5 5 2   5 2·4− ·0 2 4= 2

5 ·2−4·0 2= 2 5 2   12 5 − 2·0− 2 5 3= 2   12 3 − −8 76 5 = − 15 3

s3 :

s2 :

s:

4=

s0 :

−

135

0

0

76 ·4−0 15 . 76 − 15

We conclude that the polynomial has RHP roots, since the elements of the first column are not all positive. In fact, there are two poles in the RHP because there are two sign changes.11 Note that, in computing the Routh array, we can simplify the rest of the calculations by multiplying or dividing a row by a positive constant. Also note that the last two rows each have one nonzero element. Routh’s method is also useful in determining the range of parameters for which a feedback system remains stable.

EXAMPLE 3.31

Stability versus Parameter Range Consider the system shown in Fig. 3.37. The stability properties of the system are a function of the proportional feedback gain K. Determine the range of K over which the system is stable. Solution. The characteristic equation for the system is given by 1+K

s+1 = 0, s(s − 1)(s + 6)

or s3 + 5s2 + (K − 6)s + K = 0. Figure 3.37 A feedback system for testing stability

r






K

s1 s(s
1)(s  6)

y

11 The actual roots of the polynomial computed with the MATLAB roots command are −3.2644, 0.7797 ± 0.7488j, −0.6046 ± 0.9935j, and −0.8858, which, of course, agree with our conclusion.

136 Chapter 3 Dynamic Response The corresponding Routh array is 1 K −6 s3 : s2 : 5 K s : (4K − 30)/5 K. s0 : For the system to be stable, it is necessary that 4K − 30 >0 5

and

K > 0,

or K > 7.5

and

K > 0.

Thus, Routh’s method provides an analytical answer to the stability question. Although any gain satisfying this inequality stabilizes the system, the dynamic response could be quite different depending on the specific value of K. Given a specific value of the gain, we may compute the closed-loop poles by finding the roots of the characteristic polynomial. The characteristic polynomial has the coefficients represented by the row vector (in descending powers of s) denT = [1 5 K–6 K], Computing roots by MATLAB

and we may compute the roots using the MATLAB function roots(denT). For K = 7.5 the roots are at −5 and ±1.22j, and the system is neutrally stable. Note that Routh’s method predicts the presence of poles on the jω axis for K = 7.5. If we set K = 13, the closed-loop poles are at −4.06 and −0.47 ± 1.7j, and for K = 25, they are at −1.90 and −1.54 ± 3.27j. In both these cases, the system is stable as predicted by Routh’s method. Figure 3.38 shows the transient responses for the three gain values. To obtain these transient responses, we compute the closed-loop transfer function K(s + 1) Y (s) = 3 , T (s) = R(s) s + 5s2 + (K − 6)s + K

Figure 3.38 Transient responses for the system in Fig. 3.37

3.0 2.5

K  7.5 K  13 K  25

2.0 1.5 y

1.0 0.5 0
0.5
1.0

0

2

4

6 Time (sec)

8

10

12

3.6 Stability Figure 3.39 System with proportional-integral (PI) control

R








1 (s  1)(s  2)

KI K s

137

Y

so that the numerator polynomial is expressed as numT = [K K];

% form numerator

and denT is as before. The MATLAB commands sysT=tf(numT,denT); step(sysT);

% define system by its numerator and denominator % compute step response

produce a plot of the (unit) step response.

EXAMPLE 3.32

Stability Versus Two Parameter Ranges Find the range of the controller gains (K, KI ) so that the PI (proportional-integral; see Chapter 4) feedback system in Fig. 3.39 is stable. Solution. The characteristic equation of the closed-loop system is   1 KI = 0, 1+ K + s (s + 1)(s + 2) which we may rewrite as s3 + 3s2 + (2 + K)s + KI = 0. The corresponding Routh array is s3 s2 s s0

: 1 : 3 : (6 + 3K − KI )/3 : KI .

2+K KI

For internal stability we must have 1 KI − 2. 3 The allowable region can be plotted in MATLAB using the ensuing commands KI > 0

and

K>

fh=@(ki,k) 6+3*k−ki; ezplot(fh) hold on; f=@(ki,k) ki; ezplot(f); and is the shaded area in the (KI , K) plane shown in Fig. 3.40, which represents an analytical solution to the stability question. This example illustrates the real value of Routh’s approach and why it is superior to the numerical approaches. It would

138 Chapter 3 Dynamic Response Figure 3.40 Allowable region for stability

K

0
1
2

Figure 3.41 Transient response for the system in Fig. 3.39

6

KI

1.2 K  10, KI  5 1.0 0.8 K  1, KI  1

y 0.6 0.4

K  1, KI  0

0.2 0

0

1

2

3

4

5

6

7

8

9

10

Time (sec)

MATLAB roots

have been more difficult to arrive at these bounds on the gains using numerical search techniques. The closed-loop transfer function is Ks + KI Y (s) T (s) = = 3 . R(s) s + 3s2 + (2 + K)s + KI As in Example 3.31, we may compute the closed-loop poles for different values of the dynamic compensator gains by using the MATLAB function roots on the denominator polynomial denT = [1 3 2+K KI].

% form denominator

Similarly, we may find the zero by finding the root of the numerator polynomial numT = [K KI].

% form numerator

The closed-loop zero of the system is at −KI /K. Figure 3.41 shows the transient response for three sets of feedback gains. For K = 1 and KI = 0, the closed-loop poles are at 0 and −1.5 ± 0.86j, and there is a zero at the origin. For K = KI = 1, the poles and zeros are all at −1. For K = 10 and KI = 5, the closed-loop poles are at −0.46 and −1.26 ± 3.3j and the zero is at −0.5. The step responses were again obtained using the MATLAB function

3.6 Stability sysT=tf(numT,denT) step(sysT).

139

% define system by its numerator and denominator % compute step response

There is a large steady-state error in this case when KI = 0. (See Chapter 4.) If the first term in one of the rows is zero or if an entire row is zero, then the standard Routh array cannot be formed, so we have to use one of the special techniques described next.




Special Cases

Special case I

If only the first element in one of the rows is zero, then we can replace the zero with a small positive constant  > 0 and proceed as before. We then apply the stability criterion by taking the limit as  → 0.

EXAMPLE 3.33

Routh’s Test for Special Case I Consider the polynomial a(s) = s5 + 3s4 + 2s3 + 6s2 + 6s + 9. Determine whether any of the roots are in the RHP. Solution. The Routh array is s5 : s4 : s3 : New s3 : s2 : s: s0 :

1 3 0 

2−3 

3− 3

3 2 2−3

2 6 3 3 3 0 0

6 9 0 0 0 0

← Replace zero by 

There are two sign changes in the first column of the array, which means there are two poles not in the LHP.12

Special case II

Another special case occurs when an entire row of the Routh array is zero. This indicates that there are complex conjugate pairs of roots that are mirror images of each other with respect to the imaginary axis. If the ith row is zero, we form an auxiliary equation from the previous (nonzero) row: a1 (s) = β1 si+1 + β2 si−1 + β3 si−3 + · · · .

(3.83)

Here {βi } are the coefficients of the (i + 1)th row in the array. We then replace the ith row by the coefficients of the derivative of the auxiliary polynomial and complete 12 The actual roots computed with MATLAB are at −2.9043, 0.6567 ± 1.2881j, −0.7046 ± 0.9929j.

140 Chapter 3 Dynamic Response the array. However, the roots of the auxiliary polynomial in Eq. (3.83) are also roots of the characteristic equation, and these must be tested separately.

EXAMPLE 3.34

Routh Test for Special Case II For the polynomial a(s) = s5 + 5s4 + 11s3 + 23s2 + 28s + 12, determine whether there are any roots on the jω axis or in the RHP. Solution. The Routh array is s5 : s4 : s3 : s2 : s: New s: s0 :

1 5 6.4 3 0

11 23 25.6 12 0

6

0

28 12 0 ← a1 (s) = 3s2 + 12 ←

da1 (s) = 6s ds

12,

There are no sign changes in the first column. Hence all the roots have negative real parts except for a pair on the imaginary axis. We may deduce this as follows: When we replace the zero in the first column by  > 0, there are no sign changes. If we let  < 0, then there are two sign changes. Thus, if  = 0, there are two poles on the imaginary axis, which are the roots of a1 (s) = s2 + 4 = 0, or s = ±j2. This agrees with the fact that the actual roots are at −3, ±2j, −1, and −1, as computed using the roots command in MATLAB. The Routh–Hurwitz result assumes that the characteristic polynomial coefficients are known precisely. It is well–known that the roots of a polynomial can be very sensitive to even slight perturbations in the polynomial coefficients. If the range of variation on each one of the polynomial coefficients is known, then a remarkable result called the Kharitonov Theorem (1978) allows one to test just four so-called Kharitonov polynomials, using the Routh test, to see if the polynomial coefficient variations result in instability.


3.7 Obtaining Models from Experimental Data There are several reasons for using experimental data to obtain a model of the dynamic system to be controlled. In the first place, the best theoretical model built from equations of motion is still only an approximation of reality. Sometimes, as in the case of a very rigid spacecraft, the theoretical model is extremely good. Other times, as with many chemical processes such as papermaking or metalworking, the theoretical

3.7 Obtaining Models from Experimental Data

Four sources of experimental data

Transient response

Frequency response

Stochastic steady-state

141

model is very approximate. In every case, before the final control design is done, it is important and prudent to verify the theoretical model with experimental data. Second, in situations for which the theoretical model is especially complicated or the physics of the process is poorly understood, the only reliable information on which to base the control design is the experimental data. Finally, the system is sometimes subject to online changes, which occur when the environment of the system changes. Examples include when an aircraft changes altitude or speed, a paper machine is given a different composition of fiber, or a nonlinear system moves to a new operating point. On these occasions we need to “retune” the controller by changing the control parameters. This requires a model for the new conditions and experimental data are often the most effective, if not the only, information available for the new model. There are four kinds of experimental data for generating a model: 1. transient response, such as comes from an impulse or a step; 2. frequency-response data, which result from exciting the system with sinusoidal inputs at many frequencies; 3. stochastic steady-state information, as might come from flying an aircraft through turbulent weather or from some other natural source of randomness; 4. pseudorandom-noise data, as may be generated in a digital computer. Each class of experimental data has its properties, advantages, and disadvantages. Transient response data are quick and relatively easy to obtain. They are also often representative of the natural signals to which the system is subjected. Thus a model derived from such data can be reliable for designing the control system. On the other hand, in order for the signal-to-noise ratio to be sufficiently high, the transient response must be highly noticeable. Consequently, the method is rarely suitable for normal operations, so the data must be collected as part of special tests. A second disadvantage is that the data do not come in a form suitable for standard control systems designs, and some parts of the model, such as poles and zeros, must be computed from the data.13 This computation can be simple in special cases or complex in the general case. Frequency-response data (see Chapter 6) are simple to obtain but substantially more time consuming than transient-response information. This is especially so if the time constants of the process are large, as often occurs in chemical processing industries. As with the transient-response data, it is important to have a good signal-tonoise ratio, so obtaining frequency-response data can be very expensive. On the other hand, as we will see in Chapter 6, frequency-response data are exactly in the right form for frequency-response design methods; so once the data have been obtained, the control design can proceed immediately. Normal operating records from a natural stochastic environment at first appear to be an attractive basis for modeling systems, since such records are by definition nondisruptive and inexpensive to obtain. Unfortunately, the quality of such data is inconsistent, tending to be worst just when the control is best, because then the upsets are minimal and the signals are smooth. At such times, some or even most of the system 13 Ziegler and Nichols (1943), building on the earlier work of Callender et al. (1936), use the step response

directly in designing the controls for certain classes of processes. See Chapter 4 for details.

142 Chapter 3 Dynamic Response

Pseudorandom noise (PRBS)

dynamics are hardly excited. Because they contribute little to the system output, they will not be found in the model constructed to explain the signals. The result is a model that represents only part of the system and is sometimes unsuitable for control. In some instances, as occurs when trying to model the dynamics of the electroencephalogram (brain waves) of a sleeping or anesthetized person to locate the frequency and intensity of alpha waves, normal records are the only possibility. Usually they are the last choice for control purposes. Finally, the pseudorandom signals that can be constructed using digital logic have much appeal. Especially interesting for model making is the pseudorandom binary signal (PRBS). The PRBS takes on the value +A or −A according to the output (1 or 0) of a feedback shift register. The feedback to the register is a binary sum of various states of the register that have been selected to make the output period (which must repeat itself in finite time) as long as possible. For example, with a register of 20 bits, 220 −1 (over a million) steps are produced before the pattern repeats. Analysis beyond the scope of this text has revealed that the resulting signal is almost like a broadband random signal. Yet this signal is entirely under the control of the engineer who can set the level (A) and the length (bits in the register) of the signal. The data obtained from tests with a PRBS must be analyzed by computer and both special-purpose hardware and programs for general-purpose computers have been developed to perform this analysis.

3.7.1 Models from Transient-Response Data To obtain a model from transient data we assume that a step response is available. If the transient is a simple combination of elementary transients, then a reasonable low-order model can be estimated using hand calculations. For example, consider the step response shown in Fig. 3.42. The response is monotonic and smooth. If we assume that it is given by a sum of exponentials, we can write y(t) = y(∞) + Ae−αt + Be−βt + Ce−γ t + · · · .

(3.84)

Subtracting off the final value and assuming that −α is the slowest pole, we write y − y(∞) ∼ = Ae−αt , log10 [y − y(∞)] ∼ = log10 A − αt log10 e, ∼ = log A − 0.4343αt. 10

(3.85)

This is the equation of a line whose slope determines α and intercept determines A. If we fit a line to the plot of log10 [y−y(∞)] (or log10 [y(∞)−y] if A is negative), then we can estimate A and α. Once these are estimated, we plot y − [y(∞) + Ae−αt ], which as a curve approximates Be−βt and on the log plot is equivalent to log10 B −0.4345βt. We repeat the process, each time removing the slowest remaining term, until the data stop being accurate. Then we plot the final model step response and compare it with data so we can assess the quality of the computed model. It is possible to get a good fit to the step response and yet be far off from the true time constants (poles) of the system. However, the method gives a good approximation for control of processes whose step responses look like Fig. 3.42.

3.7 Obtaining Models from Experimental Data Figure 3.42 A step response characteristic of many chemical processes

143

y(t) 1.0

t

TABLE 3.1

Step Response Data t

y(t)

t

y(t)

0.1 0.1 0.2 0.3 0.4 0.5

0.000 0.005 0.034 0.085 0.140 0.215

1.0 1.5 2.0 2.5 3.0 4.0 ∞

0.510 0.700 0.817 0.890 0.932 0.975 1.000

Sinha and Kuszta (1983).

EXAMPLE 3.35

Determining the Model from Time–Response Data Find the transfer function that generates the data given in Table 3.1 and which are plotted in Fig. 3.43. Solution. Table 3.1 shows and Fig. 3.43 implies that the final value of the data is y(∞) = 1. We know that A is negative because y(∞) is greater than y(t). Therefore, the first step in the process is to plot log10 [y(∞) − y], which is shown in Fig. 3.44. From the line (fitted by eye) the values are log10 |A| = 0.125, 0.4343α =

0.435 1.602 − 1.167 = ⇒α∼ = 1. t 1

Thus A = −1.33, α = 1.0. If we now subtract 1 + Aeαt from the data and plot the log of the result, we find the plot of Fig. 3.45. Here we estimate log10 B = −0.48, −0.48 − (−1.7) = 2.5, 0.5 β∼ = 5.8,

0.4343β =

B = 0.33.

144 Chapter 3 Dynamic Response Figure 3.43 Step response data in Table 3.1

1.0 0.8 0.6 y(t) 0.4 0.2 0
0.2

0

1

Figure 3.44 log10 [y(∞)−y] versus t

2

3 Time (sec)

4

5

6

0.5

0.125 0


0.5 log101
y
1.0
1.167
1.5
1.602


2.0

0

1

2 Time (sec)

3

4

Combining these results, we arrive at the y estimate yˆ (t) ∼ = 1 − 1.33e−t + 0.33e−5.8t .

(3.86)

Equation (3.86) is plotted as the colored line in Fig. 3.46 and shows a reasonable fit to the data, although some error is noticeable near t = 0. From yˆ (t) we compute

3.7 Obtaining Models from Experimental Data Figure 3.45 log10 [y − (1 + Ae−αt )] versus t

145

0


0.48


0.5

log10Be
bt


1.0


1.5
1.7
2.0

0

0.1

0.2

0.3

0.4

0.5

Time (sec)

Figure 3.46 Model fits to the experimental data

1.0 0.8 0.6 y(t)

0.4 A 
1.33, a  1, B  0.33, b  5.8 0.2 A 
1.37, a  1, B  0.37, b  4.3 0 Data
0.2

0

1

2

3 Time (sec)

4

5

6

1.33 0.33 1 − + s s + 1 s + 5.8 (s + 1)(s + 5.8) − 1.33s(s + 5.8) + 0.33s(s + 1) = s(s + 1)(s + 5.8) −0.58s + 5.8 . = s(s + 1)(s + 5.8)

Yˆ (s) =

The resulting transfer function is G(s) =

−0.58(s − 10) . (s + 1)(s + 5.8)

Notice that this method has given us a system with a zero in the RHP, even though the data showed no values of y that were negative. Very small differences in the estimated value for A, all of which approximately fit the data, can cause values of

146 Chapter 3 Dynamic Response β to range from 4 to 6. This illustrates the sensitivity of pole locations to the quality of the data and emphasizes the need for a good signal-to-noise ratio. By using a computer to perform the plotting, we are better able to iterate the four parameters to achieve the best overall fit. The data presentation in Figs. 3.44 and 3.45 can be obtained directly by using a semilog plot. This eliminates having to calculate log10 and the exponential expression to find the values of the parameters. The equations of the lines to be fit to the data are y(t) = Aeαt and y(t) = Beβt , which are straight lines on a semilog plot. The parameters A and α, or B and β, are iteratively selected so that the straight line comes as close as possible to passing through the data. This process produces the improved fit shown by the dashed black line in Fig. 3.46. The revised parameters, A = −1.37, B = 0.37, and β = 4.3 result in the transfer function −0.22s + 4.3 . G(s) = (s + 1)(s + 4.3) ∼ +20 and has no noticeable The RHP zero is still present but it is now located at s = effect on the time response. This set of data was fitted quite well by a second-order model. In many cases a higher-order model is required to explain the data and the modes may not be as well separated.

Least-squares system identification

If the transient response has oscillatory modes, then these can sometimes be estimated by comparing them with the standard plots of Fig. 3.18. The period will give the frequency ωd and the decay from one period to the next will afford an estimate of the damping ratio. If the response has a mixture of modes not well separated in frequency, then more sophisticated methods need to be used. One such is leastsquares system identification, in which a numerical optimization routine selects the best combination of system parameters so as to minimize the fit error. The fit error is defined to be a scalar cost function  (ydata − ymodel )2 , i = 1, 2, 3, · · · , for each data point, J= i

so that fit errors at all data points are taken into account in determining the best value for the system parameters.

3.7.2 Models from Other Data As mentioned early in Section 3.1.2, we can also generate a model using frequencyresponse data, which are obtained by exciting the system with a set of sinusoids and plotting H(jω). In Chapter 6 we will show how such plots can be used directly for design. Alternatively, we can use the frequency response to estimate the poles and zeros of a transfer function using straight-line asymptotes on a logarithmic plot. The construction of dynamic models from normal stochastic operating records or from the response to a PRBS can be based either on the concept of cross-correlation or on the least-squares fit of a discrete equivalent model, both topics in the field of system identification. They require substantial presentation and background that are beyond the scope of this text. An introduction to system identification can be found in Chapter 8 of Franklin et al. (1998), and a comprehensive treatment is given in

3.8 Amplitude and Time Scaling

147

Ljüng (1999). Based largely on the work of Professor Ljüng, the MATLAB Toolbox on Identification provides substantial software to perform system identification and to verify the quality of the proposed models.


3.8 Amplitude and Time Scaling The magnitude of the values of the variables in a problem is often very different, sometimes so much so that numerical difficulties arise. This was a serious problem years ago when equations were solved using analog computers and it was routine to scale the variables so that all had similar magnitudes. Today’s widespread use of digital computers for solving differential equations has largely eliminated the need to scale a problem unless the number of variables is very large, because computers are now capable of accurately handling numbers with wide variations in magnitude. Nevertheless, it is wise to understand the principle of scaling for the few cases in which extreme variations in magnitude exist and scaling is necessary or the computer word size is limited.

3.8.1 Amplitude Scaling There are two types of scaling that are sometimes carried out: amplitude scaling and time scaling, as we have already seen in Section 3.1.4. Amplitude scaling is usually performed unwittingly by simply picking units that make sense for the problem at hand. For the ball levitator, expressing the motion in millimeters and the current in milliamps would keep the numbers within a range that is easy to work with. Equations of motion are sometimes developed in the standard SI units such as meters, kilograms, and amperes, but when computing the motion of a rocket going into orbit, using kilometers makes more sense. The equations of motion are usually solved using computer-aided design software, which is often capable of working in any units. For higher-order systems it becomes important to scale the problem so that system variables have similar numerical variations. A method for accomplishing the best scaling for a complex system is first to estimate the maximum values for each system variable and then to scale the system so that each variable varies between −1 and 1. In general, we can perform amplitude scaling by defining the scaled variables for each state element: If (3.87) x
= Sx x, then x˙
= Sx x˙

and

x¨
= Sx x¨ .

(3.88)

We then pick Sx to result in the appropriate scale change, substitute Eqs. (3.87) and (3.88) into the equations of motion, and recompute the coefficients.

EXAMPLE 3.36

Scaling for the Ball Levitator The linearized equation of motion for the ball levitator (see Example 9.2, Chapter 9) is δ x¨ = 1667δx + 47.6δi,

(3.89)

148 Chapter 3 Dynamic Response where δx is in units of meters and δi is in units of amperes. Scale the variables for the ball levitator to result in units of millimeters and milliamps instead of meters and amps. Solution. Referring to Eq. (3.87), we define δx
= Sx δx

and

δi
= Si δi

such that both Sx and Si have a value of 1000 in order to convert δx and δi in meters and amps to δx
and δi
in millimeters and milliamps. Substituting these relations into Eq. (3.89) and taking note of Eq. (3.88) yields δ x¨
= 1667δx
+ 47.6

Sx
δi . Si

In this case Sx = Si , so Eq. (3.89) remains unchanged. Had we scaled the two quantities by different amounts, there would have been a change in the last coefficient in the equation.

3.8.2 Time Scaling The unit of time when using SI units or English units is seconds. Computer-aided design software is usually able to compute results accurately no matter how fast or slow the particular problem at hand. However, if a dynamic system responds in a few microseconds, or if there are characteristic frequencies in the system on the order of several megahertz, the problem may become ill conditioned, so that the numerical routines produce errors. This can be particularly troublesome for high-order systems. The same holds true for an extremely slow system. It is therefore useful to know how to change the units of time should you encounter an ill-conditioned problem. We define the new scaled time to be τ = ωo t

(3.90)

such that, if t is measured in seconds and ωo = 1000, then τ will be measured in milliseconds. The effect of the time scaling is to change the differentiation so that x˙ =

dx dx dx = = ωo dt d(τ/ωo ) dτ

(3.91)

d2x d2x = ωo2 2 . 2 dt dτ

(3.92)

and x¨ =

EXAMPLE 3.37

Time Scaling an Oscillator The equation for an oscillator was derived in Example 2.5. For a case with a very fast natural frequency ωn = 15,000 rad/sec (about 2 kHz), Eq. (2.23) can be rewritten as θ¨ + 150002 · θ = 106 · Tc . Determine the time-scaled equation so that the unit of time is milliseconds.

3.9 Historical Perspective

149

Solution. The value of ωo in Eq. (3.90) is 1000. Equation (3.92) shows that d2θ ¨ = 10−6 · θ, dτ 2 and the time-scaled equation becomes d2θ + 152 · θ = Tc . dτ 2 In practice, we would then solve the equation θ¨ + 152 · θ = Tc

(3.93)

and label the plots in milliseconds instead of seconds.

3.9 Historical Perspective Oliver Heaviside (1850–1925) was an eccentric English electrical engineer, mathematician, and physicist. He was self-taught and left school at the age of 16 to become a telegraph operator. He worked mostly outside the scientific community that was hostile to him. He reformulated Maxwell’s equations in the form that is used today. He also laid down the foundations of telecommunication and hypothesized the existence of the ionosphere. He developed the symbolic procedure known as Heaviside’s operational calculus for solving differential equations. The Heaviside calculus was widely popular among electrical engineers in the 1920s and 1930s. This was later shown to be equivalent to the more rigorous Laplace transform named after the French mathematician Pierre-Simon Laplace (1749–1827) who had worked on operational calculus earlier. Laplace was also an astronomer and a mathematician who is sometimes referred to as the “The Newton of France.” He studied the origin and dynamical stability of the solar system completing Newton’s work in his five volume Méchanique céleste (Celestial Mechanics). Laplace invented the general concept of potential as in a gravitational or electric field and described by Laplace’s equation. Laplace had a brief political career as Napoleon’s Interior Minister. During a famous exchange with Napoleon who asked Laplace why he had not mentioned God in Méchanique céleste, Laplace is said to have replied that “Sir, there was no need for that hypothesis.” He was an opportunist and changed sides as the political winds shifted. Laplace’s operational property transforms a differential equation into an algebraic operation that is much easier to manipulate in engineering applications. It is also applicable to solutions of partial differential equations, the original problem that Laplace was concerned with while developing the transform. Laplace formulated the Laplace’s equation with applications to electromagnetic theory, fluid dynamics, and astronomy. Laplace also made fundamental contributions to probability theory. Laplace and Fourier transforms are intimately related (see Appendix A). The Fourier series and the Fourier transform, developed in that order, provide methods for representing signals in terms of exponential functions. Fourier series are used to represent a periodic signal with discrete spectra in terms of a series. Fourier transforms are used to represent a non-periodic signal with continuous spectra in terms of an integral. The Fourier transform is named after the French mathematician Jean Batiste Joseph

150 Chapter 3 Dynamic Response Fourier (1768–1830) who used Fourier series to solve the heat conduction equation expressed in terms of Fourier series. Laplace and Fourier were contemporaries and knew each other very well. In fact, Laplace was one of Fourier’s teachers. Fourier accompanied Napoleon on his Egyptian expedition in 1798 as a science advisor and is also credited with the discovery of the greenhouse effect. Transform methods provide a unifying method in applications to solving many engineering problems. Linear transforms such as the Laplace transform and Fourier transform are useful for studying linear systems. While Fourier transforms are useful to study the steady-state behavior, Laplace transforms are used for studying the transient and closed-loop behavior of dynamic systems. The book by Gardner and Barnes in 1942 was influential in popularizing the Laplace transform in the United States.

SUMMARY •

The Laplace transform is the primary tool used to determine the behavior of linear systems. The Laplace transform of a time function f(t) is given by

∞ L[ f (t)] = F(s) = f (t)e−st dt. (3.94) 0−

•

This relationship leads to the key property of Laplace transforms, namely, L[ f˙ (t)] = sF(s) − f (0− ).

•

•

•

(3.95)

This property allows us to find the transfer function of a linear ODE. Given the transfer function G(s) of a system and the input u(t), with transform U(s), the system output transform is Y (s) = G(s)U(s). Normally, inverse transforms are found by referring to tables such as Table A.2 in Appendix A or by computer. Properties of Laplace transforms and their inverses are summarized in Table A.1 in Appendix A. The Final Value Theorem is useful in finding steady-state errors for stable systems: If all the poles of s Y (s) are in the LHP, then lim y(t) = lim s Y (s).

t→∞

•

(3.96)

s→0

Block diagrams are a convenient way to show the relationships between the components of a system. They can usually be simplified using the relations in Fig. 3.9 and Eq. (3.50); that is, the transfer function of the block diagram R1(s)

⫹

⌺

G1(s)

Y1(s)

⫺ G2(s)

is equivalent to

G1 (s) R1 (s). (3.97) 1 + G1 (s)G2 (s) The locations of poles in the s-plane determine the character of the response, as shown in Fig. 3.15. Y1 (s) =

•

151

Review Questions •

The location of a pole in the s-plane is defined by the parameters shown in Fig. 3.22. These parameters are related to the time-domain quantities of rise time tr , settling time ts , and overshoot Mp , which are defined in Fig. 3.22. The correspondences between them, for a second-order system with no zeros, are given by 1.8 tr ∼ , = ωn

(3.98)

√ 2 Mp = e−π ζ / 1−ζ , 4.6 . ts = ζ ωn • • • •

• • •

(3.99) (3.100)

When a zero in the LHP is present, the overshoot increases. This effect is summarized in Figs. 3.26 and 3.27. When an additional stable pole is present, the system response is more sluggish. This effect is summarized in Figs. 3.34 and 3.35. For a stable system, all the closed-loop poles must be in the LHP. A system is stable if and only if all the elements in the first column of the Routh array are positive. To determine the Routh array, refer to the formulas in Section 3.6.3. Mason’s rule is a useful technique to determining transfer functions of complicated interconnected systems. Determining a model from experimental data, or verifying an analytically based model by experiment, is an important step in system design. Amplitude and time scaling (Section 3.8) are methods by which certain complications of dealing with differential equations can be minimized. Scaling of variables results in numerical values that fall within a sufficiently narrow range of magnitude to minimize errors and allow for ease of computation.

REVIEW QUESTIONS 1. What is the definition of “transfer function”? 2. What are the properties of systems whose responses can be described by transfer functions? 3. What is the Laplace transform of f (t − λ)1(t − λ) if the transform of f (t) is F(s)? 4.

State the Final Value Theorem.

5. What is the most common use of the Final Value Theorem in control? 6.

Given a second-order transfer function with damping ratio ζ and natural frequency ωn , what is the estimate of the step response rise time? What is the estimate of the percent overshoot in the step response? What is the estimate of the settling time?

7. What is the major effect of a zero in the LHP on the second-order step response? 8. What is the most noticeable effect of a zero in the RHP on the step response of the second-order system? 9. What is the main effect of an extra real pole on the second-order step response? 10. Why is stability an important consideration in control system design? 11. What is the main use of Routh’s criterion?

152 Chapter 3 Dynamic Response 12.

Under what conditions might it be important to know how to estimate a transfer function from experimental data?

PROBLEMS Problems for Section 3.1: Review of Laplace Transforms 3.1

Show that, in a partial-fraction expansion, complex conjugate poles have coefficients that are also complex conjugates. (The result of this relationship is that whenever complex conjugate pairs of poles are present, only one of the coefficients needs to be computed.)

3.2

Find the Laplace transform of the following time functions: (a) (b) (c) (d) (e)

3.3

f (t) = 1 + 2t f (t) = 3 + 7t + t 2 + δ(t) f (t) = e−t + 2e−2t + te−3t f (t) = (t + 1)2 f (t) = sinh t

Find the Laplace transform of the following time functions: (a) f (t) = 3 cos 6t (b) f (t) = sin 2t + 2 cos 2t + e−t sin 2t (c) f (t) = t 2 + e−2t sin 3t

3.4

Find the Laplace transform of the following time functions: (a) (b) (c) (d) (e)

3.5

Find the Laplace transform of the following time functions (* denotes convolution): (a) (b) (c) (d) (e)

3.6

f (t) = t sin t f (t) = t cos 3t f (t) = te−t + 2t cos t f (t) = t sin 3t − 2t cos t f (t) = 1(t) + 2t cos 2t

f (t) = sin t sin 3t f (t) = sin2 t + 3 cos2 t f (t) = (sin t)/t f (t) = sin t ∗ sin t  f (t) = 0t cos(t − τ ) sin τ dτ

Given that the Laplace transform of f (t) is F(s), find the Laplace transform of the following: (a) g(t) = f (t) cos t   (b) g(t) = 0t 0t1 f (τ ) dτ dt1

3.7

Find the time function corresponding to each of the following Laplace transforms using partial-fraction expansions: 2 (a) F(s) = s(s+2) 10 (b) F(s) = s(s+1)(s+10)

Problems (c) F(s) = (d) F(s) = (e) F(s) = (f) F(s) = (g) F(s) = (h) F(s) = (i) F(s) = (j) F(s) = 3.8

153

3s+2 s2 +4s+20 3s2 +9s+12 (s+2)(s2 +5s+11) 1 s2 +4 2(s+2) (s+1)(s2 +4) s+1 s2 1 s6 4 s4 +4 e−s s2

Find the time function corresponding to each of the following Laplace transforms: (a) F(s) = (b) F(s) = (c) F(s) = (d) F(s) = (e) F(s) = (f) F(s) = (g) F(s) =

3.9

1 s(s+2)2 2s2 +s+1 s3 −1 2(s2 +s+1) s(s+1)2 s3 +2s+4 s4 −16 2(s+2)(s+5)2 (s+1)(s2 +4)2 (s2 −1) (s2 +1)2 tan−1 ( 1s )

Solve the following ODEs using Laplace transforms: (a) (b) (c) (d) (e) (f)

y¨ (t) + y˙ (t) + 3y(t) = 0; y(0) = 1, y˙ (0) = 2 y¨ (t) − 2˙y(t) + 4y(t) = 0; y(0) = 1, y˙ (0) = 2 y¨ (t) + y˙ (t) = sin t; y(0) = 1, y˙ (0) = 2 y¨ (t) + 3y(t) = sin t; y(0) = 1, y˙ (0) = 2 y¨ (t) + 2˙y(t) = et ; y(0) = 1, y˙ (0) = 2 y¨ (t) + y(t) = t; y(0) = 1, y˙ (0) = −1

3.10 Using the convolution integral, find the step response of the system whose impulse response is given below and shown in Fig. 3.47:

Figure 3.47 Impulse response for Problem 3.10

0.4 0.35 0.3

h(t)

0.25 0.2 0.15 0.1 0.05 0 0

2

4

6

Time (sec)

8

10

154 Chapter 3 Dynamic Response 

te−t t ≥ 0, 0 t < 0. 3.11 Using the convolution integral, find the step response of the system whose impulse response is given below and shown in Fig. 3.48:  1, 0 ≤ t ≤ 2, h(t) = 0, t < 0 and t > 2. h(t) =

Figure 3.48 Impulse response for Problem 3.11

1.5

h(t)

1

0.5

0 −1

−0.5

0

0.5

1

1.5

2

2.5

3

Time (sec)

3.12 Consider the standard second-order system ωn2 . G(s) = 2 s + 2ζ ωn s + ωn2 (a) Write the Laplace transform of the signal in Fig. 3.49. (b) What is the transform of the output if this signal is applied to G(s)? (c) Find the output of the system for the input shown in Fig. 3.49.

Figure 3.49 Plot of input for Problem 3.12

u(t) 1

1

2

3

Time (sec)

3.13 A rotating load is connected to a field-controlled DC motor with negligible field inductance. A test results in the output load reaching a speed of 1 rad/sec within 1/2 sec when a constant input of 100 V is applied to the motor terminals. The output steady-state speed from the same test is found to be 2 rad/sec. Determine the transfer function Vθ (s) f (s) of the motor. 3.14 A simplified sketch of a computer tape drive is given in Fig. 3.50. (a) Write the equations of motion in terms of the parameters listed below. K and B represent the spring constant and the damping of tape stretch, respectively, and ω1 and ω2 are angular velocities. A positive current applied to the DC motor will provide a torque on the capstan in the clockwise direction as shown by the arrow. Find the value of current that just cancels the force, F, then eliminate the constant

Problems Figure 3.50 Tape drive schematic

155

B x1 No friction

x2 K

v2 r2

v1

J1, B1 r1

Take-up capstan

Head

Idler wheel J2, B2 F Vacuum column

DC motor

ia ⫽ u

current and its balancing force, F; from your equations. Assume positive angular velocities of the two wheels are in the directions shown by the arrows. J1 = 5 × 10−5 kg·m2 , motor and capstan inertia B1 = 1 × 10−2 N·m·sec, motor damping r1 = 2 × 10−2 m Kt = 3 × 10−2 N·m/A, motor–torque constant K = 2 × 104 N/m B = 20 N/m·sec r2 = 2 × 10−2 m J2 = 2 × 10−5 kg·m2 B2 = 2 × 10−2 N·m·sec, viscous damping, idler F = 6 N, constant force x˙ 1 = tape velocity m/sec (variable to be controlled) (b) Find the transfer function from the motor current to the tape position. (c) Find the poles and zeros of the transfer function in part (b). (d) Use MATLAB to find the response of x1 to a step input in ia . 3.15 For the system in Fig. 2.51, compute the transfer function from the motor voltage to position θ2 . 3.16 Compute the transfer function for the two-tank system in Fig. 2.55 with holes at A and C. 3.17 For a second-order system with transfer function 3 , G(s) = 2 s + 2s − 3

156 Chapter 3 Dynamic Response Ra

La Fm




io(t) ya(t)




ys(t) 

1:N Gear ratio

Rack and pinion

Vertically adjustable roller

Thickness T

Thickness x

Motion of material out of rollers

Fixed roller

Figure 3.51 Continuous rolling mill determine the following: (a) The DC gain; (b) The final value to a step input. 3.18 Consider the continuous rolling mill depicted in Fig. 3.51. Suppose that the motion of the adjustable roller has a damping coefficient b, and that the force exerted by the rolled material on the adjustable roller is proportional to the material’s change in thickness: Fs = c(T − x). Suppose further that the DC motor has a torque constant Kt and a back emf constant Ke , and that the rack-and-pinion has effective radius of R. (a) What are the inputs to this system? The output? (b) Without neglecting the effects of gravity on the adjustable roller, draw a block diagram of the system that explicitly shows the following quantities: Vs (s), I0 (s), F(s) (the force the motor exerts on the adjustable roller), and X(s). (c) Simplify your block diagram as much as possible while still identifying output and each input separately.

Problems for Section 3.2: System Modeling Diagrams 3.19 Consider the block diagram shown in Fig. 3.52. Note that ai and bi are constants. Compute the transfer function for this system. This special structure is called the “control canonical form” and will be discussed further in Chapter 7.

157

Problems Figure 3.52 Block diagram for Problem 3.19

b1
b2 U(s)












1 s




X1

X2

1 s

1 s


X3



Y(s)


b3

a1 a2 a3

3.20 Find the transfer functions for the block diagrams in Fig. 3.53.

Figure 3.53 Block diagrams for Problem 3.20

G2


R






G1






Y

 (a) G7 R






G3

G1






G6

G4









Y





G5

G2 (b) G7 G6 R

G1






G2




G3










G4

G5









Y

(c)

3.21 Find the transfer functions for the block diagrams in Fig. 3.54, using the ideas of block diagram simplification. The special structure in Fig. 3.54(b) is called the “observer canonical form” and will be discussed in Chapter 7.

158 Chapter 3 Dynamic Response R R








G1












G2



b3

Y




b2






H2



G3









1 s








a3

b1


1 s








a2

1 s

Y

a1


H3 (a)

(b)

R b3

b2













1 s



b1







1 s




1 s



Y

a1 a2 a3 (c)

D(s) R(s)






A(s)






B(s)









Y(s)

 H(s) G(s) (d)

Figure 3.54 Block diagrams for Problem 3.21

3.22 Use block-diagram algebra to determine the transfer function between R(s) and Y(s) in Fig. 3.55.

Problems Figure 3.55 Block diagram for Problem 3.22

159

H3


G2 R(s)




G4










H2

G1






G3



G6

Y(s)








G5 H4

Problems for Section 3.3: Effect of Pole Locations 3.23 For the electric circuit shown in Fig. 3.56, find the following: (a) The time-domain equation relating i(t) and v1 (t); (b) The time-domain equation relating i(t) and v2 (t);

2 (s) (c) Assuming all initial conditions are zero, the transfer function V V1 (s) and the damping ratio ζ and undamped natural frequency ωn of the system; (d) The values of R that will result in v2 (t) having an overshoot of no more than 25%, assuming v1 (t) is a unit step, L = 10 mH, and C = 4 µF.

Figure 3.56 Circuit for Problem 3.23

L

R







y1(t)

i(t)

C



y2(t) 

3.24 For the unity feedback system shown in Fig. 3.57, specify the gain K of the proportional controller so that the output y(t) has an overshoot of no more than 10% in response to a unit step.

Figure 3.57 Unity feedback system for Problem 3.24

R(s)




 

K

1 s(s
2)

Y(s)

3.25 For the unity feedback system shown in Fig. 3.58, specify the gain and pole location of the compensator so that the overall closed-loop response to a unit-step input has an overshoot of no more than 25%, and a 1% settling time of no more than 0.1 sec. Verify your design using MATLAB.

160 Chapter 3 Dynamic Response Figure 3.58 Unity feedback system for Problem 3.25

Compensator




R(s)

K s
a





Plant 100 s
25

Y(s)

Problems for Section 3.4: Time-Domain Specification 3.26 Suppose you desire the peak time of a given second-order system to be less than tp
. Draw the region in the s-plane that corresponds to values of the poles that meet the specification tp < tp
. 3.27 A certain servomechanism system has dynamics dominated by a pair of complex poles and no finite zeros. The time-domain specifications on the rise time (tr ), percent overshoot (Mp ), and settling time (ts ) are given by tr ≤ 0.6 sec , Mp ≤ 17%, ts ≤ 9.2 sec . (a) Sketch the region in the s-plane where the poles could be placed so that the system will meet all three specifications. (b) Indicate on your sketch the specific locations (denoted by ×) that will have the smallest rise-time and also meet the settling time specification exactly. 3.28 Suppose you are to design a unity feedback controller for a first-order plant depicted in Fig. 3.59. (As you will learn in Chapter 4, the configuration shown is referred to as a proportional-integral controller.) You are to design the controller so that the closed-loop poles lie within the shaded regions shown in Fig. 3.60. (a) What values of ωn and ζ correspond to the shaded regions in Fig. 3.59? (A simple estimate from the figure is sufficient.) (b) Let Kα = α = 2. Find values for K and KI so that the poles of the closed-loop system lie within the shaded regions.

Figure 3.59 Unity feedback system for Problem 3.28

R









e(t)






KI s

Figure 3.60 Desired closed-loop pole locations for Problem 3.28

Im(s) 4

u1 u2 4

2

2

Re(s) 2

K

Ka s
a

Y

Problems

161

(c) Prove that no matter what the values of Kα and α are, the controller provides enough flexibility to place the poles anywhere in the complex (left-half) plane. 3.29 The open-loop transfer function of a unity feedback system is G(s) =

K . s(s + 2)

The desired system response to a step input is specified as peak time tp = 1 sec and overshoot Mp = 5%.



(a) Determine whether both specifications can be met simultaneously by selecting the right value of K. (b) Sketch the associated region in the s-plane where both specifications are met, and indicate what root locations are possible for some likely values of K. (c) Relax the specifications in part (a) by the same factor and pick a suitable value for K, and use MATLAB to verify that the new specifications are satisfied. 3.30 The equations of motion for the DC motor shown in Fig. 2.32 were given in Eqs. (2.52– 2.53) as   Kt Ke Kt Jm θ¨m + b + θ˙m = va . Ra Ra Assume that Jm = 0.01 kg·m2 , b = 0.001 N·m·sec, Ke = 0.02 V·sec, Kt = 0.02 N·m/A, Ra = 10 . (a) Find the transfer function between the applied voltage va and the motor speed θ˙m . (b) What is the steady-state speed of the motor after a voltage va = 10 V has been applied? (c) Find the transfer function between the applied voltage va and the shaft angle θm . (d) Suppose feedback is added to the system in part (c) so that it becomes a position servo device such that the applied voltage is given by va = K(θr − θm ), where K is the feedback gain. Find the transfer function between θr and θm . (e) What is the maximum value of K that can be used if an overshoot Mp < 20% is desired? (f) What values of K will provide a rise time of less than 4 sec? (Ignore the Mp constraint.) (g) Use MATLAB to plot the step response of the position servo system for values of the gain K = 0.5, 1, and 2. Find the overshoot and rise time for each of the three step responses by examining your plots. Are the plots consistent with your calculations in parts (e) and (f)? 3.31 You wish to control the elevation of the satellite-tracking antenna shown in Figs. 3.61 and 3.62. The antenna and drive parts have a moment of inertia J and a damping B;

162 Chapter 3 Dynamic Response Figure 3.61 Satellite-tracking antenna Source: Courtesy Space Systems/Loral

Figure 3.62 Schematic of antenna for Problem 3.31

u

these arise to some extent from bearing and aerodynamic friction, but mostly from the back emf of the DC drive motor. The equations of motion are J θ¨ + Bθ˙ = Tc , where Tc is the torque from the drive motor. Assume that J = 600,000 kg·m2

B = 20,000 N·m·sec.

(a) Find the transfer function between the applied torque Tc and the antenna angle θ. (b) Suppose the applied torque is computed so that θ tracks a reference command θr according to the feedback law Tc = K(θr − θ ), where K is the feedback gain. Find the transfer function between θr and θ .

Problems

163

(c) What is the maximum value of K that can be used if you wish to have an overshoot Mp < 10%? (d) What values of K will provide a rise time of less than 80 sec? (Ignore the Mp constraint.) (e) Use MATLAB to plot the step response of the antenna system for K = 200, 400, 1000, and 2000. Find the overshoot and rise time of the four step responses by examining your plots. Do the plots confirm your calculations in parts (c) and (d)?



3.32 Show that the second-order system y¨ + 2ζ ωn y˙ + ωn2 y = 0,

y(0) = yo ,

y˙ (0) = 0,

has the response e−σ t y(t) = yo  sin(ωd t + cos−1 ζ ). 1 − ζ2 Prove that, for the underdamped case (ζ < 1), the response oscillations decay at a predictable rate (see Fig. 3.63) called the logarithmic decrement δ = ln = ln

yo 2π ζ = ln eσ τd = σ τd =  y1 1 − ζ2 y1 ∼ yi , = ln y1 yi

where τd =

2π 2π =  ωd ωn 1 − ζ 2

is the damped natural period of vibration. The damping coefficient in terms of the logarithmic decrement is then δ . ζ =  4π 2 + δ 2

Figure 3.63 Definition of logarithmic decrement

y0 y1

y1

y2 y2

td

t

164 Chapter 3 Dynamic Response Problems for Section 3.5: Effect of Zeros and Additional Poles 3.33 In aircraft control systems, an ideal pitch response (qo ) versus a pitch command (qc ) is described by the transfer function τ ω2 (s + 1/τ ) Qo (s) = 2 n . Qc (s) s + 2ζ ωn s + ωn2 The actual aircraft response is more complicated than this ideal transfer function; nevertheless, the ideal model is used as a guide for autopilot design. Assume that tr is the desired rise time and that 1.789 , tr 1.6 1 = , τ tr ζ = 0.89.

ωn =

Show that this ideal response possesses a fast settling time and minimal overshoot by plotting the step response for tr = 0.8, 1.0, 1.2, and 1.5 sec. 3.34 Consider the system shown in Fig. 3.64, where G(s) =

1 s(s + 3)

D(s) =

and

K(s + z) . s+p

(3.101)

Find K, z, and p so that the closed-loop system has a 10% overshoot to a step input and a settling time of 1.5 sec (1% criterion).

Figure 3.64 Unity feedback system for Problem 3.34

R(s)






D(s)

G(s)

Y(s)





3.35 Sketch the step response of a system with the transfer function G(s) =

s/2 + 1 . (s/40 + 1)[(s/4)2 + s/4 + 1]

Justify your answer on the basis of the locations of the poles and zeros. (Do not find the inverse Laplace transform.) Then compare your answer with the step response computed using MATLAB. 3.36 Consider the two nonminimum-phase systems, G1 (s) = − G2 (s) =

2(s − 1) , (s + 1)(s + 2)

3(s − 1)(s − 2) . (s + 1)(s + 2)(s + 3)

(3.102) (3.103)

(a) Sketch the unit step responses for G1 (s) and G2 (s), paying close attention to the transient part of the response. (b) Explain the difference in the behavior of the two responses as it relates to the zero locations.

Problems

165

(c) Consider a stable, strictly proper system (that is, m zeros and n poles, where m < n). Let y(t) denote the step response of the system. The step response is said to have an undershoot if it initially starts off in the “wrong”direction. Prove that a stable, strictly proper system has an undershoot if and only if its transfer function has an odd number of real RHP zeros. 3.37 Find the relationships for the impulse response and the step response corresponding to Equation (3.57) for the cases where (a) the roots are repeated. (b) the roots are both real. Express your answers in terms of hyperbolic functions (sinh, cosh) to best show the properties of the system response. (c) the value of the damping coefficient, ζ , is negative. 3.38 Consider the following second-order system with an extra pole: H(s) =

ωn2 p . (s + p)(s2 + 2ζ ωn s + ωn2 )

Show that the unit-step response is y(t) = 1 + Ae−pt + Be−σ t sin(ωd t − θ ), where −ωn2 , ωn2 − 2ζ ωn p + p2 p B=  , (p2 − 2ζ ωn p + ωn2 )(1 − ζ 2 )   1 − ζ2 ωn 1 − ζ 2 −1 −1 θ = tan + tan . −ζ p − ζ ωn A=

(a) (b) (c) (d)

Which term dominates y(t) as p gets large? Give approximate values for A and B for small values of p. Which term dominates as p gets small? (Small with respect to what?) Using the preceding explicit expression for y(t) or the step command in MATLAB, and assuming that ωn = 1 and ζ = 0.7, plot the step response of the preceding system for several values of p ranging from very small to very large. At what point does the extra pole cease to have much effect on the system response?

3.39 Consider the second-order unity DC gain system with an extra zero, H(s) =

ωn2 (s + z) . z(s2 + 2ζ ωn s + ωn2 )

(a) Show that the unit-step response for the system is given by  ω2 2ζ ωn 1 + 2n − z z e−σ t cos(ωd t + β1 ), y(t) = 1 +  2 1−ζ where

−ζ + ωzn . β1 = tan−1  1 − ζ2

166 Chapter 3 Dynamic Response (b) Derive an expression for the step response overshoot, Mp , of this system. (c) For a given value of overshoot, Mp , how do we solve for ζ and ωn ? 3.40 The block diagram of an autopilot designed to maintain the pitch attitude θ of an aircraft is shown in Fig. 3.65. The transfer function relating the elevator angle δe and the pitch attitude θ is 50(s + 1)(s + 2) θ(s) , = G(s) = 2 δe (s) (s + 5s + 40)(s2 + 0.03s + 0.06) where θ is the pitch attitude in degrees and δe is the elevator angle in degrees. The autopilot controller uses the pitch attitude error e to adjust the elevator according to the transfer function δe (s) K(s + 3) = D(s) = . E(s) s + 10 Using MATLAB, find a value of K that will provide an overshoot of less than 10% and a rise time faster than 0.5 sec for a unit-step change in θr . After examining the step response of the system for various values of K, comment on the difficulty associated with making rise time and overshoot measurements for complicated systems.

Figure 3.65 Block diagram of autopilot

ur




e

 

Control d e D(s)

Aircraft u G(s)

Problems for Section 3.6: Stability 3.41 A measure of the degree of instability in an unstable aircraft response is the amount of time it takes for the amplitude of the time response to double (see Fig. 3.66), given some nonzero initial condition. (a) For a first-order system, show that the time to double is τ2 =

ln 2 , p

where p is the pole location in the RHP. (b) For a second-order system (with two complex poles in the RHP), show that τ2 =

Figure 3.66 Time to double

ln 2 . −ζ ωn

Amplitude 2A A 0 A

Time t2

Problems

167

3.42 Suppose that unity feedback is to be applied around the listed open-loop systems. Use Routh’s stability criterion to determine whether the resulting closed-loop systems will be stable. 4(s+2) s(s3 +2s2 +3s+4) (b) KG(s) = 2(s+4) s2 (s+1) 3 +2s2 +s+1) (c) KG(s) = 4(s s2 (s3 +2s2 −s−1)

(a) KG(s) =

3.43 Use Routh’s stability criterion to determine how many roots with positive real parts the following equations have: (a) (b) (c) (d) (e)

s4 + 8s3 + 32s2 + 80s + 100 = 0. s5 + 10s4 + 30s3 + 80s2 + 344s + 480 = 0. s4 + 2s3 + 7s2 − 2s + 8 = 0. s3 + s2 + 20s + 78 = 0. s4 + 6s2 + 25 = 0.

3.44 Find the range of K for which all the roots of the following polynomial are in the LHP: s5 + 5s4 + 10s3 + 10s2 + 5s + K = 0. Use MATLAB to verify your answer by plotting the roots of the polynomial in the s-plane for various values of K. 3.45 The transfer function of a typical tape-drive system is given by G(s) =

K(s + 4) , s[(s + 0.5)(s + 1)(s2 + 0.4s + 4)]

where time is measured in milliseconds. Using Routh’s stability criterion, determine the range of K for which this system is stable when the characteristic equation is 1 + G(s) = 0. 3.46 Consider the closed-loop magnetic levitation system shown in Fig. 3.67. Determine the conditions on the system parameters (a, K, z, p, Ko ) to guarantee closed-loop system stability.

Figure 3.67 Magnetic levitation system

R








e

(s
z)

u

K(s
p)

Ko 2 (s a2)

Y

3.47 Consider the system shown in Fig. 3.68. (a) Compute the closed-loop characteristic equation. (b) For what values of (T , A) is the system stable? Hint: An approximate answer may be found using e−Ts ∼ = 1 − Ts

Figure 3.68 Control system for Problem 3.47

R








esT

A s(s
1)

Y

168 Chapter 3 Dynamic Response or e−Ts ∼ =

1 − T2 s 1 + T2 s

for the pure delay. As an alternative, you could use the computer MATLAB (SIMULINK
) to simulate the system or to find the roots of the system’s characteristic equation for various values of T and A. 3.48 Modify the Routh criterion so that it applies to the case in which all the poles are to be to the left of −α when α > 0. Apply the modified test to the polynomial s3 + (6 + K)s2 + (5 + 6K)s + 5K = 0, finding those values of K for which all poles have a real part less than −1. 3.49 Suppose the characteristic polynomial of a given closed-loop system is computed to be s4 + (11 + K2 )s3 + (121 + K1 )s2 + (K1 + K1 K2 + 110K2 + 210)s + 11K1 + 100 = 0. Find constraints on the two gains K1 and K2 that guarantee a stable closed-loop system, and plot the allowable region(s) in the (K1 , K2 ) plane. You may wish to use the computer to help solve this problem. 3.50 Overhead electric power lines sometimes experience a low-frequency, high-amplitude vertical oscillation, or gallop, during winter storms when the line conductors become covered with ice. In the presence of wind, this ice can assume aerodynamic lift and drag forces that result in a gallop up to several meters in amplitude. Large-amplitude gallop can cause clashing conductors and structural damage to the line support structures caused by the large dynamic loads. These effects in turn can lead to power outages. Assume that the line conductor is a rigid rod, constrained to vertical motion only, and suspended by springs and dampers as shown in Fig. 3.69. A simple model of this conductor galloping is

m¨y +


nπ  D(α)˙y − L(α)v y = 0, + T  (˙y2 + v2 )1/2

L
D0  0 a

Figure 3.69 Electric power-line conductor

y

2

T a

Wind y Ice

(npᐉ ) Spring constant

a

y a x

Relative wind  y2
y2 Conductor

Problems

169

where m = mass of conductor, y = conductor’s vertical displacement, D = aerodynamic drag force, L = aerodynamic lift force, v = wind velocity, α = aerodynamic angle of attack = − tan−1 (˙y/v), T = conductor tension, n = number of harmonic frequencies,  = length of conductor. Assume that L(0) = 0 and D(0) = D0 (a constant), and linearize the equation around the value y = y˙ = 0. Use Routh’s stability criterion to show that galloping can occur whenever ∂L + D0 < 0. ∂α

4

A First Analysis of Feedback

A Perspective on the Analysis of Feedback

Open-loop and closed-loop control

170

In the next three chapters we will introduce three techniques for the design of controllers. Before doing so, it is useful to develop the assumptions to be used and to derive the equations that are common to each of the design approaches we describe. As a general observation, the dynamics of systems to which control is applied are nonlinear and very complex. However, in this initial analysis, we assume that the plant to be controlled as well as the controller can be represented as dynamic systems which are linear and time invariant (LTI). We also assume that they have only single inputs and single outputs, for the most part, and may thus be represented by simple scalar transfer functions. As we mentioned in Chapter 1, our basic concerns for control are stability, tracking, regulation, and sensitivity. The goal of the analysis in this chapter is to revisit each of these requirements in a linear dynamic setting and to develop equations that will expose constraints placed on the controller and identify elementary objectives to be suggested for the controllers. The two fundamental structures for realizing controls are the open-loop structure as shown in Fig. 4.1, and the closed-loop structure, also known as feedback control, as shown in Fig. 4.2. The definition of open-loop control is that there is no closed signal path whereby the output influences the control effort. In the structure shown in Fig. 4.1, the controller transfer function modifies the reference input signal before it is applied to the plant. This controller might cancel the unwanted dynamics of the plant and replace them

4.1 The Basic Equations of Control

171

with the more desirable dynamics of the controller. In other cases open-loop control actions are taken on the plant as the environment changes, actions that are calibrated to give a good response but are not dependent on measuring the actual response. An example of this would be an aircraft autopilot whose parameters are changed with altitude or speed but not by feedback of the craft’s motion. Feedback control, on the other hand, uses a sensor to measure the output and by feedback indirectly modifies the dynamics of the system. Although it is possible that feedback may cause an otherwise stable system to become unstable (a vicious circle), feedback gives the designer more flexibility and a preferable response to each of our objectives when compared to open-loop control.

Chapter Overview The chapter begins with consideration of the basic equations of a simple open-loop structure and of an elementary feedback structure. In Section 4.1 the equations for the two structures are presented in general form and compared in turn with respect to stability, tracking, regulation, and sensitivity. In Section 4.2 the steady-state errors in response to polynomial inputs are analyzed in more detail. As part of the language of steady-state performance, control systems are assigned a type number according to the maximum degree of the input polynomial for which the steady-state error is a finite constant. For each type an appropriate error constant is defined, which allows the designer to easily compute the size of this error. Although Maxwell and Routh developed a mathematical basis for assuring stability of a feedback system, design of controllers from the earliest days was largely trial and error based on experience. From this tradition there emerged an almost universal controller, the proportional–integral–derivative (PID) structure considered in Section 4.3. This device has three elements: a Proportional term to close the feedback loop, an Integral term to assure zero error to constant reference and disturbance inputs, and a Derivative term to improve (or realize!) stability and good dynamic response. In this section these terms are considered and their respective effects illustrated. As part of the evolution of the PID controller design, a major step was the development of a simple procedure for selecting the three parameters, a process called “tuning the controller.” Ziegler and Nichols developed and published a set of experiments to be run, characteristics to be measured, and tuning values to be recommended as a result. These procedures are discussed in this section. Finally, in optional Section 4.4, a brief introduction to the increasingly common digital implementation of controllers is given. Sensitivity of time response to parameter changes is discussed in Appendix W4 on the web.

4.1 The Basic Equations of Control We begin by collecting a set of equations and transfer functions that will be used throughout the rest of the text. For the open-loop system of Fig. 4.1, if we take the disturbance to be at the input of the plant, the output is given by

172 Chapter 4 A First Analysis of Feedback Figure 4.1 Open-loop system showing reference, R, control, U, disturbance, W, and output Y Figure 4.2 Closed-loop system showing the reference, R, control, U, disturbance, W, output, Y , and sensor noise, V

W(s) R(s)

Controller Dol (s)




u(s)


Plant G(s)



Y(s)

W(s) R(s)




Controller Dcl (s)

 









Y(s)

Plant G(s)







Yol = GDol R + GW

V

(4.1)

and the error, the difference between reference input and system output, is given by Eol = R − Yol

(4.2)

= R − [GDol R + GW ]

(4.3)

= [1 − GDol ]R − GW .

(4.4)

The open-loop transfer function in this case is Tol (s) = G(s)Dol (s). For feedback control, Fig. 4.2 gives the basic unity feedback structure of interest. There are three external inputs: the reference, R, which the output is expected to track, the plant disturbance, W , which the control is expected to counteract so it does not disturb the output, and the sensor noise, V , which the controller is supposed to ignore. For the feedback block diagram of Fig. 4.2, the equations for the output and the control are given by the superposition of the responses to the three inputs individually, as follows: GDcl G GDcl R+ W− V. (4.5) Ycl = 1 + GDcl 1 + GDcl 1 + GDcl Dcl GDcl Dcl U= R− W− V. (4.6) 1 + GDcl 1 + GDcl 1 + GDcl Perhaps more important than these is the equation for the error, Ecl = R − Ycl .   GDcl G GDcl Ecl = R − R+ W− V (4.7) 1 + GDcl 1 + GDcl 1 + GDcl =

G GDcl 1 R− W+ V. 1 + GDcl 1 + GDcl 1 + GDcl

(4.8)

GDcl . 1 + GDcl With these equations we will explore the four basic objectives of stability, tracking, regulation, and sensitivity for both the open-loop and the closed-loop cases. In this case, the closed-loop transfer function is Tcl =

173

4.1 The Basic Equations of Control

4.1.1 Stability As we saw in Chapter 3, the requirement for stability is simply stated: All poles of the transfer function must be in the left half-plane (LHP). In the open-loop case described by Eq. (4.1), these are the poles of GDol . To see the restrictions this requirement places on the controller, we define the polynomials a(s), b(s), c(s), and d(s) so that b(s) c(s) bc and Dol (s) = d(s) . Therefore GDol = ad . With these definitions, the G(s) = a(s) stability requirement is that neither a(s) nor d(s) may have roots in the right half-plane (RHP). A naive engineer might believe that if the plant is unstable with a(s) having a root in the RHP, the system might be made stable by canceling this pole with a zero of c(s). However, the unstable pole remains and the slightest noise or disturbance will cause the output to grow until stopped by saturation or system failure. Likewise, if the plant shows poor response because of a zero of b(s) in the RHP, an attempt to fix this by cancellation using a root of d(s) will similarly result in an unstable system. We conclude that an open-loop structure cannot be used to make an unstable plant to be stable and therefore cannot be used if the plant is already unstable. For the feedback system, from Eq. (4.8), the system poles are the roots of 1 + GDcl = 0. Again using the polynomials defined above, the system characteristic equation is 1 + GDcl = 0 b(s)c(s) 1+ =0 a(s)d(s) a(s)d(s) + b(s)c(s) = 0.

(4.9) (4.10) (4.11)

From this equation, it is clear that the feedback case grants considerably more freedom to the controller design than does the open-loop case. However, one must still avoid unstable cancellations. For example, if the plant is unstable and therefore a(s) has a root in the RHP, we might cancel this pole by putting a zero of c(s) at the same place. However, Eq. (4.11) shows that as a result, the unstable pole remains a pole of the system and this method will not work. However, unlike the open-loop case, having a pole of a(s) in the RHP does NOT prevent our designing a feedback controller that will make the system stable. For example, in Chapter 2 we derived the transfer function for the inverted pendulum, which, for simple values, might be G(s) = s21−1 for which we have b(s) = 1 and a(s) = s2 −1 = (s +1)(s −1). Suppose

we try D(s) =

K(s+γ ) s+δ .

The characteristic equation that results for the system is

(s + 1)(s − 1)(s + δ) + K(s + γ ) = 0.

(4.12)

This is the problem that Maxwell faced in his study of governors, namely under what conditions on the parameters will all the roots of this equation be in the LHP? The problem was solved by Routh. In our case, a simple solution is to take γ = 1 and the common (stable) factor cancels. The resulting second-order equation can be easily solved to place the remaining two poles at any point desired. Exercise. If we wish to force the characteristic equation to be s2 + 2ζ ωs + ω2 = 0, solve for K and δ in terms of ζ and ω.

174 Chapter 4 A First Analysis of Feedback 4.1.2 Tracking The tracking problem is to cause the output to follow the reference input as closely as possible. In the open-loop case, if the plant is stable and has neither poles nor zeros in the RHP, then in principle the controller can be selected to cancel the transfer function of the plant and substitute whatever desired transfer function the engineer wishes. This apparent freedom, however, comes with three caveats. First, in order to physically build it, the controller transfer function must be proper, meaning that it cannot be given more zeros than it has poles. Second, the engineer must not get greedy and request an unrealistically fast design. This entire analysis has been based on the assumption that the plant is linear and a demand for a fast response will demand large inputs to the plant, inputs that will be sure to saturate the system if the demand is too great. Again, it is the responsibility of the engineer to know the limits of the plant and to set the desired overall transfer function to a reasonable value with this knowledge. Third and finally, although one can, in principle, stably cancel any pole in the LHP, the next section on sensitivity faces up to the fact that the plant transfer function is subject to change and if one tries to cancel a pole that is barely inside the LHP there is a good chance of disaster as that pole moves a bit and exposes the system response to unacceptable transients. 1 it is proposed to use a conExercise. For a plant having the transfer function s2 +3s+9 +c1 s+c0 . Solve troller in a unity feedback system and having the transfer function c2 ss(s+d 1) for the parameters of this controller so that the closed-loop will have the characteristic equation (s + 6)(s + 3)(s2 + 3s + 9) = 01 . {ans: c2 = 34, c1 = 36, c0 = 162, d1 = 11} 2

Exercise. Show that if the reference input to the system of the above exercise is a step of amplitude A, the steady-state error will be zero.

4.1.3 Regulation The problem of regulation is to keep the error small when the reference is at most a constant set point and disturbances are present. A quick look at the open-loop block diagram reveals that the controller has no influence at all on the system response to either of the disturbances, w, or v, so this structure is useless for regulation. We turn to the feedback case. From Eq. (4.8) we find a conflict between w and v in the search for a good controller. For example, the term giving the contribution of the plant disturbance G W . To select Dcl to make this term small, we should make to the system error is 1+GD cl Dcl as large as possible and infinite if that is feasible. On the other hand, the error GDcl V . In this case, unfortunately, if we select Dcl to term for the sensor noise is 1+GD cl be large, the transfer function tends to unity and the sensor noise is not reduced at all! What are we to do? The resolution of the dilemma is to observe that each of these terms is a function of frequency so one of them can be large for some frequencies and small for others. With this in mind, we also note that the frequency content of most plant disturbances occurs at very low frequencies and in fact, the most common case is a bias, which is all at zero frequency! On the other hand, a good sensor will have 1 This process is called “pole placement,” a technique to be discussed in Chapter 7.

4.1 The Basic Equations of Control

175

no bias and can be constructed to have very little noise over the entire range of low frequencies of interest. Thus, using this information, we design the controller transfer function to be large at the low frequencies, where it will reduce the effect of w, and we make it small at the higher frequencies, where it will reduce the effects of the high frequency sensor noise. The control engineer must determine in each case the best place on the frequency scale to make the cross over from amplifying to attenuation. Exercise. Show that if w is a constant bias and if Dcl has a pole at s = 0 then the error due to this bias will be zero. However, show that if G has a pole at zero, it does not help with a disturbance bias.

4.1.4 Sensitivity Suppose a plant is designed with gain G at a particular frequency but in operation it changes to be G + δG. This represents a fractional or percent change of gain of δG/G. For the purposes of this analysis, we set the frequency at zero and take the open-loop controller gain to be fixed at Dol (0). In the open-loop case the nominal overall gain is thus Tol = GDol , and with the perturbed plant gain, the overall gain would be Tol + δTol = Dol (G + δG) = Dol G + Dol δG = Tol + Dol δG. Therefore, the gain change is δTol = Dol δG. The sensitivity, SGT , of a transfer function, Tol , to a plant gain, G, is defined to be the ratio of the fractional change in Tol defined ol as δT Tol to the fractional change in G. In equation form δTol T SGT = ol δG G G δTol . = Tol δG

(4.13)

(4.14)

Substituting the values, we find that δG Dol δG δTol = . = Tol Dol G G

(4.15)

This means that a 10% error in G would yield a 10% error in Tol . In the open-loop case, therefore, we have computed that S = 1. From Eq. (4.5), the same change in G in the feedback case yields the new steadystate feedback gain as Tcl + δTcl =

(G + δG)Dcl , 1 + (G + δG)Dcl

where Tcl is the closed-loop gain. We can compute the sensitivity of this closed-loop gain directly using differential calculus. The closed-loop steady-state gain is Tcl =

GDcl . 1 + GDcl

176 Chapter 4 A First Analysis of Feedback The first-order variation is proportional to the derivative and is given by dTcl δG. dG The general expression for sensitivity from Eq. (4.13) is given by δTcl =

SGTcl  sensitivity of Tcl with respect to G, SGTcl 

G dTcl , Tcl dG

(4.16)

so (1 + GDcl )Dcl − Dcl (GDcl ) G GDcl /(1 + GDcl ) (1 + GDcl )2 1 = . 1 + GDcl

SGTcl =

Advantage of feedback

(4.17)

This result exhibits a major advantage of feedback:2 In feedback control, the error in the overall transfer function gain is less 1 compared sensitive to variations in the plant gain by a factor of S = 1+DG to errors in open-loop control gain. If the gain is such that 1 + DG = 100, a 10% change in plant gain G will cause only a 0.1% change in the steady-state gain. The open-loop controller is 100 times more sensitive to gain changes than the closed-loop system with loop gain of 100. The example of the unity feedback case is so common that we will refer to the result of Eq. (4.17) simply as the sensitivity, S, without subscripts or superscripts. The results in this section so far have been computed under the assumption of the steady-state error in the presence of constant inputs, either reference or disturbance. Very similar results can be obtained for the steady-state behavior in the presence of a sinusoidal reference or disturbance signal. This is important because there are times when such signals naturally occur as, for example, with a disturbance of 60 Hertz due to power-line interference in an electronic system. The concept is also important because more complex signals can be described as containing sinusoidal components over a band of frequencies and analyzed using superposition of one frequency at a time. For example, it is well known that human hearing is restricted to signals in the frequency range of about 60 to 15,000 Hertz. A feedback amplifier and loudspeaker system designed for high-fidelity sound must accurately track any sinusoidal (pure tone) signal in this range. If we take the controller in the feedback system shown in Fig. 4.2 to have the transfer function D(s) and we take the process to have the transfer function G(s), then the steady-state open-loop gain at the sinusoidal signal of frequency ωo will be |G(jωo )D(jωo )| and the error of the feedback system will be     1 .  (4.18) |E(jωo )| = |R(jωo )|  1 + G(jωo )D(jωo )  2 Bode, who developed the theory of sensitivity as well as many other properties of feedback, defined

sensitivity as S = 1 + GD, the inverse of our choice.

177

4.1 The Basic Equations of Control

Thus, to reduce errors to 1% of the input at the frequency ωo , we must make |1+DG| ≥ 100 or, effectively, |D(jωo )G(jωo )|  100 and a good audio amplifier must have this loop gain over the range 2π60 ≤ ω ≤ 2π 15000. We will revisit this concept in Chapter 6 as part of the design based on frequency response techniques.

The Filtered Case Thus far the analysis has been based on the simplest open- and closed-loop structures. A more general case includes a dynamic filter on the input and also dynamics in the sensor. The filtered open-loop structure is shown in Fig. 4.3 having the transfer function Tol = GDol F. In this case, the open-loop controller transfer function has been simply replaced by DF and the discussion given for the unfiltered open-loop case is easily applied to this change. For the filtered feedback case shown in Fig. 4.4, the changes are more siginificant. In that case, the transform of the system output is given by Y=

G HGDcl GDcl F R+ W− V. 1 + GDcl H 1 + GDcl H 1 + GDcl H

(4.19)

As is evident from this equation, the sensor dynamics, H is part of the loop transfer function and enters into the question of stability with Dcl H replacing the Dcl of the unity feedback case. In fact, if F = H then, with respect to stability, tracking, and regulation, the filtered case is identical to the unity case with Dcl H replacing Dcl . On the other hand, the filter transfer function F can play the role of the open-loop controller except that here the filter F would be called on to modify the entire loop transfer funcGDcl tion, 1+GD , rather than simply GDol . Therefore the filtered closed-loop structure cl H can realize the best properties of both the open-loop and the unity feedback closedloop cases. The controller, Dcl , can be designed to effectively regulate the system for the disturbance W and the sensor noise, V , while the filter F is designed to improve the tracking accuracy. If the sensor dynamics, H, are accessible to the designer, this term can also be designed to improve the response to the sensor noise. The remaining issue is sensitivity.

Figure 4.3 Filtered open-loop system

Figure 4.4 Filtered closed-loop. R = reference, u = control, Y = output, and V = sensor noise

W(s)

R(s)

Filter F (s )

Rf (s)

Controller Dol (s)

u(s)







Plant G (s )



Y(s)

W R(s)

Filter F(s)

Rf (s)


m(s)

 

Controller Dcl (s)




u(s)






Sensor H(s)

Y(s)

Plant G(s)









V(s)

178 Chapter 4 A First Analysis of Feedback Using the formula given in Eq. (4.13), with changes in the parameter of interest, we can compute SFTcl = 1.0, 1 , 1 + GDcl H GDcl H . = 1 + GDcl H

(4.20)

SGTcl =

(4.21)

SHTcl

(4.22)

Of these, the most interesting is the last. Notice that with respect to H, the sensitivity approaches unity as the loop gain grows. Therefore it is particularly important that the transfer function of the sensor be not only low in noise but also very stable in gain. Money spent on the sensor is money well spent.

4.2 Control of Steady-State Error to Polynomial Inputs: System Type In studying the regulator problem, the reference input is taken to be a constant. It is also the case that the most common plant disturbance is a constant bias. Even in the general tracking problem the reference input is often constant for long periods of time or may be adequately approximated as if it were a polynomial in time, usually one of low degree. For example, when an antenna is tracking the elevation angle to a satellite, the time history as the satellite approaches overhead is an S-shaped curve as sketched in Fig. 4.5. This signal may be approximated by a linear function of time (called a ramp function or velocity input) for a significant time relative to the speed of response of the servomechanism. As another example, the position control of an elevator has a ramp function reference input, which will direct the elevator to move with constant speed until it comes near the next floor. In rare cases, the input can even be approximated over a substantial period as having a constant acceleration. Consideration of these cases leads us to consider steady-state errors in stable systems with polynomial inputs. As part of the study of steady-state errors to polynomial inputs, a terminology has been developed to express the results. For example, we classify systems as to “type” according to the degree of the polynomial that they can reasonably track. For example, a system that can track a polynomial of degree 1 with a constant error is called Type 1. Also, to quantify the tracking error, several “error constants” are defined. In all of the following analysis, it is assumed that the systems are stable, else the analysis makes no sense at all.

Figure 4.5 Signal for satellite tracking

us

Time (sec)

4.2 Control of Steady-State Error to Polynomial Inputs: System Type

179

4.2.1 System Type for Tracking In the unity feedback case shown in Fig. 4.2, the system error is given by Eq. (4.8). If we consider tracking the reference input alone and set W = V = 0, then the equation for the error is simply E=

1 R = SR. 1 + GDcl

(4.23)

To consider polynomial inputs, we let r(t) = t /k!1(t) for which the transform is 1 R = sk+1 . We take a mechanical system as the basis for a generic reference nomenclature, calling step inputs for which k = 0 “position” inputs, ramp inputs for which k = 1 are called “velocity” inputs and if k = 2, the inputs are called “acceleration” inputs, regardless of the units of the actual signals. Application of the Final Value Theorem to the error formula gives the result k

lim e(t) = ess = lim sE(s)

t→∞

s→0

1 R(s) 1 + GDcl 1 1 . = lim s s→0 1 + GDcl sk+1 = lim s s→0

(4.24) (4.25) (4.26)

We consider first a system for which GDcl has no pole at the origin and a step input for which R(s) = 1/s. Thus r(t) is a polynomial of degree 0. In this case, Eq. (4.26) reduces to 1 1 s→0 1 + GDcl s 1 . = 1 + GDcl (0)

ess = lim s

(4.27) (4.28)

We define this system to be Type 0 and we define the constant, GDcl (0)  Kp as the “position error constant.” Notice that if the input should be a polynomial of degree higher than 1, the resulting error would grow without bound. A polynomial of degree 0 is the highest degree a system of Type 0 can track at all. If GDcl (s) has one pole at the origin, we could continue this line of argument and consider first-degree polynomial inputs but it is quite straightforward to evaluate Eq. (4.26) in a general setting. For this case, it is necessary to describe the behavior of the controller and plant as s approaches 0. For this purpose, we collect all the terms except the pole(s) at the origin into a function GDclo (s), which is finite at s = 0 so that we can define the constant GDclo (0) = Kn and write the loop transfer function as GDcl (s) =

GDclo (s) . sn

(4.29)

180 Chapter 4 A First Analysis of Feedback For example, if GDcl has no integrator, then n = 0. If the system has one integrator, then n = 1, and so forth. Substituting this expression into Eq. (4.26), 1 1 GDclo (s) sk+1 1+ sn n s 1 . = lim n s→0 s + Kn sk

ess = lim s

(4.30)

s→0

(4.31)

From this equation we can see at once that if n > k then e = 0 and if n < k then 1 e → ∞. If n = k = 0, then ess = 1+K and if n = k = 0, then ess = 1/Kn . As we saw 0 above, if n = k = 0, the input is a zero-degree polynomial otherwise known as a step or position, the constant Ko is called the “position constant” written as Kp , and the system is classified as “Type 0.” If n = k = 1, the input is a first-degree polynomial otherwise known as a ramp or velocity input and the constant K1 is called the “velocity constant” written as Kv . This system is classified “Type 1” (read “type one”). In a similar way, systems of Type 2 and higher types may be defined. A clear picture of the situation is given by the plot in Fig. 4.6 for a system of Type 1 having a ramp reference input. The error between input and output of size K1v is clearly marked. Using Eq (4.29), these results can be summarized by the equations:

Kp = lim GDcl (s),

n = 0,

(4.32)

Kv = lim sGDcl (s),

n = 1,

(4.33)

Ka = lim s2 GDcl (s),

n = 2.

(4.34)

s→0 s→0 s→0

The type information can also be usefully gathered in a table of error values as a function of the degree of the input polynomial and the type of the system as shown in Table 4.1. Figure 4.6 Relationship between ramp response and Kv

10 9 8 7 6 r, y

ess 

5

r

y

1 Ky

4 3 2 1 0

0

1

2

3

4

5 6 Time (sec)

7

8

9

10

4.2 Control of Steady-State Error to Polynomial Inputs: System Type

TABLE 4.1

Errors as a Function of System Type Type Input Type 0

EXAMPLE 4.1

181

Step (position)

Ramp(velocity)

Parabola (acceleration)

1 1 + Kp

∞

∞ ∞

Type 1

0

1 Kv

Type 2

0

0

1 Ka

System Type for Speed Control Determine the system type and the relevant error constant for speed control with A proportional feedback given by D(s) = kp . The plant transfer function is G = τ s+1 . k A

p and applying Eq. (4.32) we see that n = 0 in Solution. In this case, GDcl = τ s+1 this case as there is no pole at s = 0. Thus the system is Type 0 and the error constant is a position constant given by Kp = kp A.

EXAMPLE 4.2

System Type Using Integral Control Determine the system type and the relevant error constant for the speed control example with proportional plus integral control having controller given by Dc = kp + kI/s. A The plant transfer function is G = τ s+1 . A(kp s+kI ) Solution. In this case, the loop transfer function is GDcl (s) = s(τ s+1) and, as a unity feedback system with a single pole at s = 0, the system is immediately seen as Type 1. The velocity constant is given by Eq. (4.33) to be Kv = lim sGDcl (s) = AkI . s→0

Robustness of system type

The definition of system type helps us to identify quickly the ability of a system to track polynomials. In the unity feedback structure, if the process parameters change without removing the pole at the origin in a Type 1 system, the velocity constant will change but the system will still have zero steady-state error in response to a constant input and will still be Type 1. Similar statements can be made for systems of Type 2 or higher. Thus we can say that system type is a robust property with respect to parameter changes in the unity feedback structure. Robustness is a major reason for preferring unity feedback over other kinds of control structure. Another form of the formula for the error constants can be developed directly in terms of the closed-loop transfer function T (s). From Fig. 4.4 the transfer function including a sensor transfer function is Y (s) GD = T (s) = , R(s) 1 + GDH and the system error is E(s) = R(s) − Y (s) = R(s) − T (s)R(s).

(4.35)

182 Chapter 4 A First Analysis of Feedback The reference-to-error transfer function is thus E(s) = 1 − T (s), R(s) and the system error transform is E(s) = [1 − T (s)]R(s). We assume that the conditions of the Final Value Theorem are satisfied, namely that all poles of sE(s) are in the LHP. In that case the steady-state error is given by applying the Final Value Theorem to get ess = lim e(t) = lim sE(s) = lim s[1 − T (s)]R(s). t→∞

s→0

s→0

(4.36)

If the reference input is a polynomial of degree k, the error transform becomes E(s) =

1 sk+1

[1 − T (s)]

and the steady-state error is given again by the Final Value Theorem: 1 − T (s) 1 − T (s) = lim . (4.37) k+1 s→0 s→0 s sk As before, the result of evaluating the limit in Eq. (4.37) can be zero, a nonzero constant, or infinite and if the solution to Eq. (4.37) is a nonzero constant, the system is referred to as Type k. Notice that a system of Type 1 or higher has a closed-loop DC gain of 1.0, which means that T (0) = 1 in these cases. ess = lim s

EXAMPLE 4.3

System Type for a Servo with Tachometer Feedback Consider an electric motor position control problem including a non-unity feedback system caused by having a tachometer fixed to the motor shaft and its voltage (which is proportional to shaft speed) is fed back as part of the control. The parameters are 1 , s(τ s + 1) D(s) = kp , G(s) =

H(s) = 1 + kt s, F(s) = 1. Determine the system type and relevant error constant with respect to reference inputs. Solution. The system error is E(s) = R(s) − Y (s) = R(s) − T (s)R(s) DG(s) R(s) 1 + HDG(s) 1 + (H(s) − 1)DG(s) = R(s). 1 + HDG(s) = R(s) −

4.2 Control of Steady-State Error to Polynomial Inputs: System Type

183

The steady-state system error from Eq. (4.37) is ess = lim sR(s)[1 − T (s)]. s→0

For a polynomial reference input, R(s) = 1/sk+1 and hence ess = lim

s→0

[1 − T (s)] 1 s(τ s + 1) + (1 + kt s − 1)kp = lim k s→0 s s(τ s + 1) + (1 + kt s)kp sk

= 0, =

1 + kt kp , kp

k=0 k = 1; k

therefore the system is Type 1 and the velocity constant is Kv = 1+kpt kp . Notice that if kt > 0, perhaps to improve stability or dynamic response, the velocity constant is smaller than with simply the unity feedback value of kp . The conclusion is that if tachometer feedback is used to improve dynamic response, the steady-state error is usually increased.

4.2.2 System Type for Regulation and Disturbance Rejection A system can also be classified with respect to its ability to reject polynomial disturbance inputs in a way analogous to the classification scheme based on reference inputs. The transfer function from the disturbance input W (s) to the error E(s) is E(s) −Y (s) = = Tw (s) W (s) W (s)

(4.38)

because, if the reference is equal to zero, the output is the error. In a similar way as for reference inputs, the system is Type 0 if a step disturbance input results in a nonzero constant steady-state error and is Type 1 if a ramp disturbance input results in a steady-state value of the error that is a non zero constant, etc. In general, following the same approach used in developing Eq. (4.31), we assume that a constant n and a function To,w (s) can be defined with the properties that To,w (0) = 1/Kn,w and that the disturbance-to-error transfer function can be written as Tw (s) = sn To,w (s).

(4.39)

Then the steady-state error to a disturbance input, which is a polynomial of degree k, is   1 yss = lim sTw (s) k+1 s→0 s   sn = lim To,w (s) k . (4.40) s→0 s From Eq. (4.40), if n > k, then the error is zero and if n < k, the error is unbounded. If n = k, the system is type k and the error is given by 1/Kn,w .

184 Chapter 4 A First Analysis of Feedback Figure 4.7 DC motor with unity feedback

W(s)

B A

R










D(s)




A s(ts
1)



Y

1.0

EXAMPLE 4.4

System Type for a DC Motor Position Control Consider the simplified model of a DC motor in unity feedback as shown in Fig. 4.7, where the disturbance torque is labeled W (s). This case was considered in Example 2.11. (a) Use the controller (4.41) D(s) = kp , and determine the system type and steady-state error properties with respect to disturbance inputs. (b) Let the controller transfer function be given by kI (4.42) D(s) = kp + , s and determine the system type and the steady-state error properties for disturbance inputs. Solution. (a) The closed-loop transfer function from W to E (where R = 0) is −B Tw (s) = s(τ s + 1) + Akp = s0 To,w , n = 0, −Akp Ko,w = . B Applying Eq. (4.40) we see that the system is Type 0 and the steady-state error to a unit step torque input is ess = −B/Akp . From the earlier section, this system is seen to be Type 1 for reference inputs and illustrates that system type can be different for different inputs to the same system. (b) For this controller the disturbance error transfer function is −Bs Tw (s) = 2 , (4.43) s (τ s + 1) + (kp s + kI )A n = 1, AkI Kn,w = , −B

(4.44) (4.45)

4.2 Control of Steady-State Error to Polynomial Inputs: System Type

185

and therefore the system is Type 1 and the error to a unit ramp disturbance input will be −B ess = . (4.46) AkI

Truxal’s Formula for the Error Constants Truxal (1955) derived a formula for the velocity constant of a Type 1 system in terms of the closed-loop poles and zeros, a formula that connects the steady-state error to the system’s dynamic response. Since control design often requires a trade-off between these two characteristics, Truxal’s formula can be useful to know. Its derivation is quite direct. Suppose the closed-loop transfer function T (s) of a Type 1 system is (s − z1 )(s − z2 ) · · · (s − zm ) . (4.47) (s − p1 )(s − p2 ) · · · (s − pn ) Since the steady-state error in response to a step input in a Type 1 system is zero, the DC gain is unity; thus T (0) = 1. (4.48) T (s) = K

The system error is given by



 Y (s) = R(s)[1 − T (s)]. E(s)  R(s) − Y (s) = R(s) 1 − R(s) The system error due to a unit ramp input is given by E(s) =

1 − T (s) . s2

(4.49)

(4.50)

Using the Final Value Theorem, we get 1 − T (s) . (4.51) s→0 s Using L’Hôpital’s rule we rewrite Eq. (4.51) as dT ess = − lim (4.52) s→0 ds or 1 dT ess = − lim = . (4.53) s→0 ds Kv Equation (4.53) implies that 1/Kv is related to the slope of the transfer function at the origin, a result that will also be shown in Section 6.1.2. Using Eq. (4.48), we can rewrite Eq. (4.53) as dT 1 ess = − lim (4.54) s→0 ds T or d (4.55) ess = − lim [ln T (s)]. s→0 ds Substituting Eq. (4.47) into Eq. (4.55), we get   m  d (s − zi ) i=1 ess = − lim ln K n (4.56) s→0 ds i=1 (s − pi ) ess = lim

186 Chapter 4 A First Analysis of Feedback

  m m   d ln(s − zi ) − ln(s − pi ) = − lim K+ s→0 ds

or

i=1

(4.57)

i=1

 n m  d ln T  1 1 1 =− = − + . Kv ds s=0 pi zi i=1

(4.58)

i=1

We observe from Eq. (4.58) that Kv increases as the closed-loop poles move away from the origin. Similar relationships exist for other error coefficients, and these are explored in the problems.

EXAMPLE 4.5

Truxal’s Formula

Truxal’s formula

A third-order Type 1 system has closed-loop poles at −2 ± 2j and −0.1. The system has only one closed-loop zero. Where should the zero be if a Kv = 10 is desired? Solution. From Truxal’s formula we have, 1 1 1 1 1 − − + =− Kv −2 + 2j −2 − 2j −0.1 z or 1 0.1 = 0.5 + 10 + , z 1 = 0.1 − 0.5 − 10, z = −10.4 Therefore, the closed-loop zero should be at z = 1/−10.4 = −0.0962.

4.3 The Three-Term Controller: PID Control In later chapters we will study three general analytic and graphical design techniques based on the root locus, the frequency response, and the state space formulation of the equations. Here we describe a control method having an older pedigree that was developed through long experience and by trial and error. Starting with simple proportional feedback, engineers early discovered integral control action as a means of eliminating bias offset. Then, finding poor dynamic response in many cases, an “anticipatory” term based on the derivative was added. The result is called the threeterm or PID controller and has the transfer function3 kI + kD s, (4.59) s where kp is the proportional term, kI is the integral term, and kD is the derivative term. We’ll discuss them in turn. D(s) = kp +

3 The derivative term alone makes this transfer function nonproper and impractical. However adding a

high-frequency pole to make the term proper only slightly modifies the performance.

4.3 The Three-Term Controller: PID Control

187

4.3.1 Proportional Control (P) When the feedback control signal is linearly proportional to the system error, we call the result proportional feedback. This was the case for the feedback used in the controller of speed in Section 4.1 for which the controller transfer function is U(s) = Dcl (s) = kp . (4.60) E(s) If the plant is second order, as, for example, for a motor with nonnegligible inductance, then the plant transfer function can be written as A G(s) = 2 . (4.61) s + a1 s + a 2 In this case, the characteristic equation with proportional control is 1 + kp G(s) = 0,

(4.62)

s2 + a1 s + a2 + kp = 0.

(4.63)

The designer can control the constant term in this equation, which determines the natural frequency, but cannot control the damping of the equation. The system is Type 0 and if kp is made large to get adequately small steady-state error, the damping may be much too low for satisfactory transient response with proportional control alone.

4.3.2 Proportional Plus Integral Control (PI) Proportional plus integral control

Adding an integral term to the controller to get the automatic reset effect results in the proportional plus integral control equation in the time domain:

t u(t) = kp e + kI e(τ ) dτ , (4.64) t0

for which the Dcl (s) in Fig. 4.2 becomes U(s) kI = Dcl (s) = kp + . (4.65) E(s) s Introduction of the integral term raises the type to Type 1 and the system can therefore reject completely constant bias disturbances. For example, consider PI control in a speed control example, where the plant is described by A (U + W ). (4.66) Y= τs + 1 The transform equation for the controller is R−Y U = kp (R − Y ) + kI , (4.67) s and the system transform equation with this controller is   kI (R − Y ) + AW , (4.68) (τ s + 1)Y = A kp + s and, if we multiply by s and collect terms, (τ s2 + (Akp + 1)s + AkI )Y = A(kp s + kI )R + sAW .

(4.69)

188 Chapter 4 A First Analysis of Feedback Because the PI controller includes dynamics, use of this controller will change the dynamic response. This we can understand by considering the characteristic equation given by (4.70) τ s2 + (Akp + 1)s + AkI = 0. The two roots of this equation may be complex and, if so, the natural frequency is  Ak +1 ωn = AkI/τ and the damping ratio is ζ = 2τpωn . These parameters may both be determined by the controller gains. On the other hand, if the plant is second order, described by A , (4.71) G(s) = 2 s + a1 s + a 2 then the characteristic equation of the system is kp s + k I A = 0, 2 s s + a1 s + a2

(4.72)

s3 + a1 s2 + a2 s + Akp s + AkI = 0.

(4.73)

1+

In this case, the controller parameters can be used to set two of the coefficients but not the third. For this we need derivative control.

4.3.3 PID Control The final term in the classical controller is derivative control, D. An important effect of this term is that it gives a sharp response to suddenly changing signals. Because of this, the “D” term is sometimes introduced into the feedback path as shown in Fig. 4.8(a). This could be either a part of the standard controller or could describe a velocity sensor such as a tachometer on the shaft of a motor. The closed-loop characteristic equation is the same as if the term were in the forward path as given by Eq. (4.59) and drawn in Fig. 4.8(b). It is important to notice that the zeros from the reference to the output are Figure 4.8 Block diagram of the PID controller: (a) with the D-term in the feedback path; and (b) with the D-term in the forward path

W

R






kP




kI s

U







G(s)

Y

 kDs

(a)

W

R




 

kP


U kI
kDs s


(b)






G(s)

Y

4.3 The Three-Term Controller: PID Control

189

different in the two cases. With the derivative in the feedback path, the reference is not differentiated, which is how the undesirable response to sudden changes is avoided. To illustrate the effect of a derivative term on PID control, consider speed control but with the second-order plant. In that case, the characteristic equation is s2 + a1 s + a2 + A(kp +

kI + kD s) = 0, s

s3 + a1 s2 + a2 s + A(kp s + kI + kD s2 ) = 0.

(4.74)

Collecting terms results in s3 + (a1 + AkD )s2 + (a2 + Akp )s + AkI = 0.

(4.75)

The point here is that this equation, whose three roots determine the nature of the dynamic response of the system, has three free parameters in kp , kI , and kD and that by selection of these parameters, the roots can be uniquely and, in theory, arbitrarily determined. Without the derivative term, there would be only two free parameters, but with three roots, the choice of roots of the characteristic equation would be restricted. To illustrate the effect more concretely, a numerical example is useful.

EXAMPLE 4.6

PID Control of Motor Speed Consider the DC motor speed control with parameters4 Jm = 1.13 × 10−2 b = 0.028 N·m·sec/rad, La = 10−1 henry, 2 N·m· sec /rad, Kt = 0.067 N·m/amp, Ke = 0.067 volt·sec/rad Ra = 0.45 ohms, (4.76) These parameters were defined in Example 2.11 in Chapter 2. Use the controller parameters kp = 3,

kI = 15 sec,

kD = 0.3 sec.

(4.77)

Discuss the effects of P, PI, and PID control on the responses of this system to steps in the disturbance torque and steps in the reference input. Let the unused controller parameters be zero. Solution. Figure 4.9(a) illustrates the effects of P, PI, and PID feedback on the step disturbance response of the system. Note that adding the integral term increases the oscillatory behavior but eliminates the steady-state error and that adding the derivative term reduces the oscillation while maintaining zero steady-state error. Figure 4.9(b) illustrates the effects of P, PI, and PID feedback on the step reference response with similar results. The step responses can be computed by forming the numerator and denominator coefficient vectors (in descending powers of s) and using the step function in MATLAB.


4 These values have been scaled to measure time in milliseconds by multiplying the true L and J by 1000 a m

each.

190 Chapter 4 A First Analysis of Feedback 1.8

8

1.6

6

P

PI

1.2

2

Amplitude

Amplitude

4 PI

0

1.0 0.8 0.6

2

P

0.4

PID

4 6

PID

1.4

0.2 0

1

2

3 4 Time (msec)

5

6

0

0

1

2

3 4 Time (msec)

5

6

(b)

(a)

Figure 4.9 Responses of P, PI, and PID control to (a) step disturbance input (b) step reference input

EXAMPLE 4.7

PI Control for a DC Motor Position Control Consider the simplified model of a DC motor in unity feedback as shown in Fig. 4.7 where the disturbance torque is labeled W (s). Let the sensor be −h rather than −1. (a) Use the proportional controller D(s) = kp

(4.78)

and determine the system type and steady-state error properties with respect to disturbance inputs. (b) Let the control be PI as given by kI (4.79) s and determine the system type and the steady-state error properties for disturbance inputs. D(s) = kp +

Solution. (a) The closed-loop transfer function from W to E (where R = 0) is Tw (s) =

−B s(τ s + 1) + Akp h

= s0 To,w , n = 0, −Akp h Ko,w = . B Applying Eq. (4.40) we see that the system is Type 0 and the steady-state error to a unit step torque input is ess = −B/Akph . From the earlier section, this system is seen to be Type 1 for reference inputs and illustrates that system type can be different for

4.3 The Three-Term Controller: PID Control

191

different inputs to the same system. However, in this case the system is Type 0 for reference inputs. (b) If the controller is PI, the disturbance error transfer function is Tw (s) =

−Bs , s2 (τ s + 1) + (kp s + kI )Ah

(4.80)

n = 1, (4.81) AkI h Kn,w = , (4.82) −B and therefore the system is Type 1 and the error to a unit ramp disturbance input in this case will be −B . (4.83) ess = AkI h

EXAMPLE 4.8

Satellite Attitude Control Consider the model of a satellite attitude control system shown in Fig. 4.10 (a) where J = moment of inertia, W = disturbance torque, K = sensor and reference gain, D(s) = the compensator. With equal input filter and sensor scale factors, the system with PD control can be redrawn with unity feedback as in Fig. 4.10(b) and with PID control drawn as in Fig. 4.10(c). Assume that the control results in a stable system and determine the system types and error responses to disturbances of the control system for (a) System Fig. 4.10(b) Proportional plus derivative control where D(s) = kP + k D s (b) System Fig. 4.10(c) Proportional plus integral plus derivative control where D = kp + kI/s + kD s.5 Solution. (a) We see from inspection of Fig. 4.10(b) that with two poles at the origin in the plant, the system is Type 2 with respect to reference inputs. The transfer function from disturbance to error is 1 Tw (s) = 2 (4.84) Js + kD s + kp = To,w (s)

(4.85)

for which n = 0 and Ko,w = kp . The system is Type 0 and the error to a unit disturbance step is 1/kp . 5 Notice that these controller transfer functions have more zeros than poles and are therefore not practical. In practice, the derivative term would have a high-frequency pole, which has been omitted for simplicity in these examples.

192 Chapter 4 A First Analysis of Feedback Figure 4.10 Model of a satellite attitude control: (a) basic system; (b) PD control; (c) PID control

W


K

R



D(s)



U





u

1 Js



1 s

uY

K

(a) W





R



kp
kDs







1 Js2



Y

1.0

(b) W

R







kp
kI
kDs s







1 Js2

Y

1.0

(c)

(b) With PID control, the forward gain has three poles at the origin, so this system is Type 3 for reference inputs but the disturbance transfer function is s , (4.86) Tw (s) = 3 Js + kD s2 + kp s + kI n = 1, To,w (s) =

1 Js3 + kD s2 + kp s + kI

(4.87) (4.88)

from which the system is Type 1 and the error constant is kI ; so the error to a disturbance ramp of unit slope will be 1/kI .

4.3.4 Ziegler–Nichols Tuning of the PID Controller When the PID controller was being developed, selecting values for the several terms (known as “tuning” the controller) was often a hit and miss affair. To bring order to the situation and make life easier for plant operators, control engineers looked for

4.3 The Three-Term Controller: PID Control Figure 4.11 Process reaction curve

193

y(t) A Slope R  t  Reaction rate A t

A

L  td Lag

Transfer function for a high-order system with a characteristic process reaction curve

Tuning by decay ratio of 0.25

t

t

ways to make the tuning more systematic. Callender et al. (1936) proposed a design for PID controllers by specifying satisfactory values for the terms based on estimates of the plant parameters that an operating engineer could make from experiments on the process itself. This approach was extended by J. G. Ziegler and N. B. Nichols (1942, 1943) who recognized that the step responses of a large number of process control systems exhibit a process reaction curve like that shown in Fig. 4.11, which can be generated from experimental step response data. The S-shape of the curve is characteristic of many systems and can be approximated by the step response of a plant with transfer function Ae−std Y (s) = , (4.89) U(s) s which is a first-order system with a time delay or “transportation lag” of td sec. The constants in Eq. (4.89) can be determined from the unit step response of the process. If a tangent is drawn at the inflection point of the reaction curve, then the slope of the line is R = A/τ , the intersection of the tangent line with the time axis identifies the time delay L = td and the final value gives the value of A.6 Ziegler and Nichols gave two methods for tuning the PID controller for such a model. In the first method the choice of controller parameters is designed to result in a closed-loop step response transient with a decay ratio of approximately 0.25. This means that the transient decays to a quarter of its value after one period of oscillation, as shown in Fig. 4.12. A quarter decay corresponds to ζ = 0.21 and, while low for many applications, was seen as a reasonable compromise between quick response and adequate stability margins for the process controls being considered. The authors simulated the equations for the system on an analog computer and adjusted the controller parameters until the transients showed the decay of 25% in one period. The regulator parameters suggested by Ziegler and Nichols for the controller terms defined by 1 Dc (s) = kp (1 + + TD s) (4.90) TI s are given in Table 4.2. 6 K. J. Astrom and others have pointed out that a time constant, τ , can also be estimated from the curve

and claim that a more effective tuning can be done by including that parameter.

194 Chapter 4 A First Analysis of Feedback Figure 4.12 Quarter decay ratio

y(t) 1

Period

0.25

t

TABLE 4.2

Ziegler–Nichols Tuning for the Regulator D(s) = K(1 + 1/TI s + TD s), for a Decay Ratio of 0.25 Type of Controller

Optimum Gain

P

kp = 1/RL  kp = 0.9/RL TI = L/0.3 ⎧ ⎨ kp = 1.2/RL TI = 2L ⎩ TD = 0.5L

PI

PID

Figure 4.13 Determination of ultimate gain and period

Tuning by evaluation at limit of stability (ultimate sensitivity method)

r






e

y Ku

Process



In the ultimate sensitivity method the criteria for adjusting the parameters are based on evaluating the amplitude and frequency of the oscillations of the system at the limit of stability rather than on taking a step response. To use the method, the proportional gain is increased until the system becomes marginally stable and continuous oscillations just begin with amplitude limited by the saturation of the actuator. The corresponding gain is defined as Ku (called the ultimate gain) and the period of oscillation is Pu (called the ultimate period). These are determined as shown in Figs. 4.13 and 4.14. Pu should be measured when the amplitude of oscillation is as small as possible. Then the tuning parameters are selected as shown in Table 4.3. Experience has shown that the controller settings according to Ziegler–Nichols rules provide acceptable closed-loop response for many systems. The process operator will often do final tuning of the controller iteratively on the actual process to yield satisfactory control.

4.3 The Three-Term Controller: PID Control Figure 4.14 Neutrally stable system

195

y(t)

t

Pu

TABLE 4.3

Ziegler–Nichols Tuning for the Regulator Dc (s) = kp (1 + 1/TI s + TD s), Based on the Ultimate Sensitivity Method Type of Controller

Optimum Gain

P

kp = 0.5Ku ⎧ ⎨ kp = 0.45Ku P ⎩ TI = u 1.2 ⎧ ⎨ kp = 1.6Ku T = 0.5Pu ⎩ I TD = 0.125Pu

PI

PID

Figure 4.15 A measured process reaction curve

1.2 1.0 0.8 y 0.6 0.4 0.2 0 0.0

EXAMPLE 4.9

100.0

200.0 Time (sec)

300.0

400.0

Tuning of a Heat Exchanger: Quarter Decay Ratio Consider the heat exchanger discussed in Chapter 2. The process reaction curve of this system is shown in Fig. 4.15. Determine proportional and PI regulator gains for the system using the Zeigler–Nichols rules to achieve a quarter decay ratio. Plot the corresponding step responses.

196 Chapter 4 A First Analysis of Feedback 1.8

1.8

1.6

1.6

1.4

1.4 PI

PI

1.2

1.2

1.0

1.0

y

y 0.8 0.6

0.8 Proportional

0.6

0.4

0.4

0.2

0.2

0 0.0

100.0

200.0 Time (sec)

300.0

400.0

(a)

0 0.0

Proportional

100.0

200.0 Time (sec)

300.0

400.0

(b)

Figure 4.16 Closed-loop step responses

Solution. From the process reaction curve we measure the maximum slope to be 1 and the time delay to be L ∼ R∼ = 13 sec. According to the Zeigler–Nichols rules = 90 of Table 4.2 the gains are 90 1 = = 6.92, RL 13 13 0.9 L = 6.22 and TI = = = 43.3. PI : kp = RL 0.3 0.3

Proportional : kp =

Figure 4.16(a) shows the step responses of the closed-loop system to these two regulators. Note that the proportional regulator results in a steady-state offset, while the PI regulator tracks the step exactly in the steady state. Both regulators are rather oscillatory and have considerable overshoot. If we arbitrarily reduce the gain kp by a factor of 2 in each case, the overshoot and oscillatory behaviors are substantially reduced, as shown in Fig. 4.16(b).

EXAMPLE 4.10

Tuning of a Heat Exchanger: Oscillatory Behavior Proportional feedback was applied to the heat exchanger in the previous example until the system showed nondecaying oscillations in response to a short pulse (impulse) input, as shown in Fig. 4.17. The ultimate gain is measured to be Ku = 15.3, and the period was measured at Pu = 42 sec. Determine the proportional and PI regulators according to the Zeigler–Nichols rules based on the ultimate sensitivity method. Plot the corresponding step responses.

4.3 The Three-Term Controller: PID Control Figure 4.17 Ultimate period of heat exchanger

197

0.010 0.008

Impulse response

0.006 0.004 0.002 0.00 0.002 0.004 0.006 0.008 0.010

0

20

40

60 80 Time (sec)

1.8

1.8

1.6

1.6

1.4

100

120

1.4

PI

1.2

1.2

1.0

1.0

y

PI

y 0.8

0.8 Proportional

0.6

0.6

0.4

0.4

0.2

0.2

0 0.0

100.0

200.0 Time (sec)

300.0

400.0

Proportional

0 0.0

100.0

200.0 Time (sec)

(a)

300.0

400.0

(b)

Figure 4.18 Closed-loop step response Solution. The regulators from Table 4.3 are Proportional : kp = 0.5,

Ku = 7.65,

1 Pu = 35. 1.2 The step responses of the closed-loop system are shown in Fig. 4.18(a). Note that the responses are similar to those in Example 4.9. If we reduce kp by 50%, then the overshoot is substantially reduced, as shown in Fig. 4.18(b). PI : kp = 0.45,

Ku = 6.885

and

TI =

198 Chapter 4 A First Analysis of Feedback

4.4 Introduction to Digital Control As a result of the revolution in the cost-effectiveness of digital computers, there has been an increasing use of digital logic in embedded applications such as controllers in feedback systems. A digital controller gives the designer much more flexibility to make modifications to the control law after the hardware design is fixed because the formula for calculating the control signal is in the software rather than the hardware. In many instances, this means that the hardware and software designs can proceed almost independently, saving a great deal of time. Also, it is relatively easy to include binary logic and nonlinear operations as part of the function of a digital controller as compared to an analog controller. Special processors designed for real-time signal processing and known as digital signal processors (DSPs) are particularly well suited for use as real-time controllers. Chapter 8 includes a more extensive introduction to the math and concepts associated with the analysis and design of digital controllers and digital control systems. However, in order to be able to compare the analog designs of the next three chapters with reasonable digital equivalents, we give here a brief introduction to the most simple techniques for digital designs. A digital controller differs from an analog controller in that the signals must be sampled and quantized.7 A signal to be used in digital logic needs to be sampled first and then the samples need to be converted by an analog-to-digital converter or A/D8 into a quantized digital number. Once the digital computer has calculated the proper next control signal value, this value needs to be converted back into a voltage and held constant or otherwise extrapolated by a digital-to-analog converter or D/A9 in order to be applied to the actuator of the process. The control signal is not changed until the next sampling period. As a result of sampling, there are strict limits on the speed and bandwidth of a digital controller. Discrete design methods that tend to minimize these limitations are described in Chapter 8, which tend to minimize these limitations. A reasonable rule of thumb for selecting the sampling period is that during the risetime of the response to a step, the input to the discrete controller should be sampled approximately six times. By adjusting the controller for the effects of sampling, the sample period can be as large as two to three times per rise time. This corresponds to a sampling frequency that is 10 to 20 times the system’s closed-loop bandwidth. The quantization of the controller signals introduces an equivalent extra noise into the system and to keep this interference at an acceptable level, the A/D converter usually has an accuracy of 10 to 12 bits although inexpensive systems have been designed with only 8 bits. For a first analysis, the effects of the quantization are usually ignored, as they will be in this introduction. A simplified block diagram of a system with a digital controller is shown in Fig. 4.19. For this introduction to digital control, we will describe a simplified technique for finding a discrete (sampled but not quantized) equivalent to a given continuous controller. The method depends on the sampling period, Ts , being short enough that the reconstructed control signal is close to the signal that the original analog controller 7A controller that operates on signals that are sampled but not quantized is called discrete while one that

operates on signals that are both sampled and quantized is called digital. 8 Pronounced “A to D.” 9 Often spelled DAC and pronounced as one word to rhyme with quack.

199

4.4 Introduction to Digital Control Figure 4.19 Block diagram of a digital controller

R






A/D

e(kT)



Digital controller u(kT) D(z)

T

D/A

U

Plant G

Y

Clock Sensor H

would have produced. We also assume that the numbers used in the digital logic have enough accurate bits so that the quantization implied in the A/D and D/A processes can be ignored. While there are good analysis tools to determine how well these requirements are met, here we will test our results by simulation, following the well-known advice that “The proof of the pudding is in the eating.” Finding a discrete equivalent to a given analog controller is equivalent to finding a recurrence equation for the samples of the control, which will approximate the differential equation of the controller. The assumption is that we have the transfer function of an analog controller and wish to replace it with a discrete controller that will accept samples of the controller input, e(kTs ), from a sampler and, using past values of the control signal, u(kTs ) and present and past samples of the input, e(kTs ) will compute the next control signal to be sent to the actuator. As an example, consider a PID controller with the transfer function kI (4.91) U(s) = (kp + + kD s)E(s), s which is equivalent to the three terms of the time-domain expression

t u(t) = kp e(t) + kI e(τ )dτ + kD e˙ (t) (4.92) 0

= uP + uI + uD .

(4.93)

Based on these terms and the fact that the system is linear, the next control sample can be computed term-by-term. The proportional term is immediate: uP (kTs + Ts ) = kp e(kTs + Ts ).

(4.94)

The integral term can be computed by breaking the integral into two parts and approximating the second part, which is the integral over one sample period, as follows.

kTs +Ts

uI (kTs + Ts ) = kI

e(τ )dτ

0 kTs

= kI 0

(4.95)

e(τ )dτ + kI

kTs +Ts

e(τ )dτ

(4.96)

kTs

= uI (kTs ) + {area under e(τ ) over one period} (4.97) Ts ∼ (4.98) = uI (kTs ) + kI {e(kTs + Ts ) + e(kTs )}. 2 In Eq. (4.98) the area in question has been approximated by that of the trapezoid formed by the base Ts and vertices e(kTs + Ts ) and e(kTs ) as shown by the dashed line in Fig. 4.20.

200 Chapter 4 A First Analysis of Feedback Figure 4.20 Graphical interpretation of numerical integration

x

x  f (x, u) x(ti)

 x dt t

0

0

Trapezoid rule

ti

ti
1

t

The area can also be approximated by the rectangle of amplitude e(kTs ) and width Ts shown by the solid blue in Fig. 4.20 to give uI (kTs + Ts ) = uI (kTs ) + kI Ts e(kTs ). These and other possibilities are considered in Chapter 8. In the derivative term, the roles of u and e are reversed from integration and the consistent approximation can be written down at once from Eq. (4.98) and Eq. (4.92) as Ts (4.99) {uD (kTs + Ts ) + uD (kTs )} = kD {e(kTs + Ts ) − e(kTs )}. 2 As with linear analog transfer functions, these relations are greatly simplified and generalized by the use of transform ideas. At this time, the discrete transform will be introduced simply as a prediction operator z much as if we described the Laplace transform variable, s, as a differential operator. Here we define the operator z as the forward shift operator in the sense that if U(z) is the transform of u(kTs ) then zU(z) will be the transform of u(kTs + Ts ). With this definition, the integral term can be written as Ts (4.100) zUI (z) = UI (z) + kI [zE(z) + E(z)] , 2 Ts z + 1 E(z), (4.101) UI (z) = kI 2 z−1 and from Eq. (4.99), the derivative term becomes the inverse as 2 z−1 UD (z) = kD E(z). (4.102) Ts z + 1 The complete discrete PID controller is thus described by   Ts z + 1 2 z−1 U(z) = kp + kI + kD E(z). (4.103) 2 z−1 Ts z + 1 Comparing the two discrete equivalents of integration and differentiation with the corresponding analog terms, it is seen that the effect of the discrete approximation in the z domain is as if everywhere in the analog transfer function, the operator s 10 has been replaced by the composite operator T2s z−1 z+1 . This is the trapezoid rule of discrete equivalents: The discrete equivalent to Da (s) is   2 z−1 (4.104) Dd (z) = Da Ts z + 1 10 The formula is also called Tustin’s Method after the English engineer who used the technique to study the responses of nonlinear circuits.

4.4 Introduction to Digital Control

EXAMPLE 4.11

201

Discrete Equivalent Find the discrete equivalent to the analog controller having transfer function D(s) = using the sample period Ts = 1. Solution. The discrete operator is

11s + 1 U(s) = E(s) 3s + 1

2(z−1) z+1

(4.105)

and thus the discrete transfer function is

U(z) = D(s)| 2 z − 1 E(z) s= Ts z + 1  2(z−1) 11 z+1 + 1 . =  +1 3 2(z−1) z+1

Dd (z) =

(4.106)

(4.107)

Clearing fractions, the discrete transfer function is Dd (z) =

23z − 21 U(z) = . E(z) 7z − 5

(4.108)

Converting the discrete transfer function to a discrete difference equation using the definition of z as the forward shift operator is done as follows. First we cross-multiply in Eq. (4.108) to obtain (7z − 5)U(z) = (23z − 21)E(z)

(4.109)

and, interpreting z as a shift operator, this is equivalent to the difference equation11 7u(k + 1) − 5u(k) = 23e(k + 1) − 21e(k),

(4.110)

where we have replaced kTs + Ts , with k + 1 to simplify the notation. To compute the next control at time kTs + Ts , therefore, we solve the difference equation u(k + 1) =

23 21 5 u(k) + e(k + 1) − e(k). 7 7 7

(4.111)

Now let’s apply these results to a control problem. Fortunately MATLAB
provides us with the SIMULINK
capability to simulate both continuous and discrete systems allowing us to compare the responses of the systems with continuous and discrete controllers.

EXAMPLE 4.12

Equivalent Discrete Controller for Speed Control A motor speed control is found to have the plant transfer function 45 Y = . U (s + 9)(s + 5)

(4.112)

11 The process is entirely similar to that used in Chapter 3 to find the ordinary differential equation to which a rational Laplace transform corresponds.

202 Chapter 4 A First Analysis of Feedback A PI controller designed for this system has the transfer function s+6 U = 1.4 . (4.113) E s The closed-loop system has a rise time of about 0.2 sec and an overshoot of about 20%. Design a discrete equivalent to this controller and compare the step responses and control signals of the two systems. (a) Compare the responses if the sample period is 0.07, which is about three samples per rise time. (b) Compare the responses with a sample period of Ts = 0.035, which corresponds to about six samples per rise time. Solution. (a) Using the substitution given by Eq. (4.104), the discrete equivalent for 2 z−1 Ts = 0.07 is given by replacing s by s ← 0.07 z+1 in D(s) as follows: D(s) =

2 z−1 +6 Dd (z) = 1.4 0.07 z + 1 2 z−1 0.07 z + 1 2(z − 1) + 6 ∗ 0.07(z + 1) = 1.4 2(z − 1) 1.21z − 0.79 = 1.4 . (z − 1)

(4.114)

(4.115) (4.116)

Based on this expression, the equation for the control is (the sample period is suppressed) u(k + 1) = u(k) + 1.4 ∗ [1.21e(k + 1) − 0.79e(k)].

(4.117)

(b) For Ts = 0.035, the discrete transfer function is Dd = 1.4

1.105z − 0.895 , z−1

(4.118)

for which the difference equation is u(k + 1) = u(k) + 1.4[1.105 e(k + 1) − 0.895 e(k)]. A SIMULINK block diagram for simulating the two systems is given in Fig. 4.21 and plots of the step responses are given in Fig. 4.22(a). The respective control signals are plotted in Fig. 4.22(b). Notice that the discrete controller for Ts = 0.07 results in a substantial increase in the overshoot in the step response while with Ts = 0.035 the digital controller matches the performance of the analog controller fairly well. For controllers with many poles and zeros, making the continuous-to-discrete substitution called for in Eq. (4.104) can be very tedious. Fortunately, MATLAB provides a command that does all the work. If one has a continuous transfer function given by Dc (s) = numD denD represented in MATLAB as sysDa = tf(numD,denD), then the discrete equivalent with sampling period Ts is given by sysDd = c2d(sysDa, Ts , ‘t’).

(4.119)

In this expression, of course, the polynomials are represented in MATLAB form. The last parameter in the c2d function given by ‘t’ calls for the conversion to be done

4.5 Historical Perspective Tau 1 9 s
9

Slider Kc 1.4

PI Control s
6 s


 Step

Tau 2 5 s
5

Mux

Mux1

Control

Tau 1 9 s
9

Slider Kd 1.4

Discrete 1.21z0.79 PI control z1




203

Output

Tau 2 5 s
5

Figure 4.21 SIMULINK
block diagram to compare continuous and discrete controllers. 1.4

2.5

Digital controller (T  0.07 sec)

1.2

Continuous controller

2.0

1.0

Digital controller (T  0.07 sec) 1.5

0.8

Continuous controller

Discrete controller (T  0.035 sec) 0.6

1.0 Discrete controller (T  0.035 sec)

0.4 0.5 0.2 0

0

0.5

1

1.5

2 2.5 3 Time (sec)

3.5

4

4.5

5

0

0

0.5

1

(a)

1.5

2 2.5 3 Time (sec)

3.5

4

4.5

5

(b)

Figure 4.22 Comparison plots of a speed control system with continuous and discrete controllers: (a) output responses. (b) control signals using the trapezoid method. The alternatives can be found by asking MATLAB for help c2d. For example, to compute the polynomials for Ts = 0.07 for the example above, the commands would be numDa = [1 6]; denDa = [1 - 0]; sysDa = tf(numD,denD) sysDd = c2d( sysDa,0.07,‘t’)

4.5 Historical Perspective The field of control is characterized by two paths: theory and practice. Control theory is basically the application of mathematics to solve control problems while control

204 Chapter 4 A First Analysis of Feedback practice, as used here, is the practical application of feedback in devices where it is found to be useful. Historically, practical applications have come first with control being introduced by trial and error. Although the applicable mathematics is often known, the theory describing how the control works and pointing the way to improvements has typically been applied later. For example, James Watt’s company began manufacturing steam engines using the fly-ball governor in 1788 but it was not until 1840 that G. B. Airy described instability in a similar device and not until 1868 than J. C. Maxwell published “On Governors” with a theoretical description of the problem. Then it was not until 1877, almost 100 years after the steam engine control was introduced, that E. J. Routh published a solution giving the requirements for stability. This situation has been called the “Gap between Theory and Practice” and continues to this day as a source of creative tension that stimulates both theory and practice. Regulation is central to the process industries, from making beer to making gasoline. In these industries there are a host of variables that need to be kept constant. Typical examples are temperature, pressure, volume, flow rates, composition, and chemical properties such as pH level. However, before one can regulate by feedback, one must be able to measure the variable of interest and before there was control there were sensors. In 1851, George Taylor and David Kendall founded the company that later became the Taylor Instrument Company in Rochester, NY, to make thermometers and barometers for weather forecasting. In 1855 they were making thermometers for several industries, including the brewing industry where they were used for manual control. Other early entries into the instrument field were the Bristol Company, founded in Naugatuck, CT, in 1889 by William Bristol, and the Foxboro Company, founded in Foxboro, MA, in 1908 by William’s father and two of his brothers. For example, one of Bristol’s instruments was used by Henry Ford to measure (and presumably control) steam pressure while he worked at the Detroit Edison Company. The Bristol Company pioneered in telemetry that permitted instruments to be placed at a distance from the process so a plant manager could monitor several variables at once. As the instruments became more sophisticated, and devices such as motordriven valves became available, they were used in feedback control often using simple on–off methods as described in Chapter 1 for the home furnace. An important fact was that the several instrument companies agreed upon standards for the variables used so a plant could mix and match instruments and controllers from different suppliers. In 1920 Foxboro introduced a controller based on compressed air that included reset or integral action. Eventually, each of these companies introduced instruments and controllers that could implement full PID action. A major step was taken for tuning PID controllers in 1942 when Ziegler and Nichols, working for Taylor Instruments, published their method for tuning based on experimental data. The poster child for the tracking problem was that of the anti-aircraft gun, whether on land or at sea. The idea was to use radar to track the target and to have a controller that would predict the path of the aircraft and aim the gun to a position such that the projectile would hit the target when it got there. The Radiation Laboratory was set up at MIT during World War II to develop such radars, one of which was the SCR-584. Interestingly, one of the major contributors to the control methods developed for this project was none other than Nick Nichols who had earlier worked on tuning PID controllers. When the record of the Rad Lab was written, Nichols was selected to be one of the editors of volume 25 on control.

Summary

205

H. S. Black joined Bell Laboratories in 1921 and was assigned to find a design for an electronic amplifier suitable for use as a repeater on the long lines of the telephone company. The basic problem was that the gain of the vacuum tube components he had available drifted over time and he needed a design that, over the audio frequency range, maintained a specific gain with great precision in the face of these drifts. Over the next few years he tried many approaches, including a feed-forward technique designed to cancel the tube distortion. While this worked in the laboratory, it was much too sensitive to be practical in the field. Finally, in August of 1927,12 while on the ferry boat from Staten Island to Manhattan, he realized that negative feedback might work and he wrote the equations on the only paper available, a page of the New York Times. He applied for a patent in 1928 but it was not issued until December 1937.13 The theory of sensitivity and many other theories of feedback were worked out by H. W. Bode.

SUMMARY • •

• •

The most important measure of the performance of a control system is the system error to all inputs. Compared to open-loop control, feedback can be used to stabilize an otherwise unstable system, to reduce errors to plant disturbances, to improve the tracking of reference inputs and to reduce the system’s transfer function sensitivity to parameter variations. Sensor noise introduces a conflict between efforts to reduce the error caused by plant disturbances and efforts to reduce the errors caused by the sensor noise. Classifying a system as Type k indicates the ability of the system to achieve zero steady-state error to polynomials of degree less than but not equal to k. A stable unity feedback system is Type k with respect to reference inputs if the loop gain G(s)D(s) has k poles at the origin in which case we can write G(s)D(s) =

A(s + z1 )(s + z2 ) · · · sk (s + p1 )(s + p2 ) · · ·

and the error constant is given by Kk = lim sk G(s)D(s) s→0

• •

(4.120)

A table of steady-state errors for unity feedback systems of Types 0, 1, and 2 to reference inputs is given in Table 4.1. Systems can be classified as to type for rejecting disturbances by computing the system error to polynomial disturbance inputs. The system is Type k to disturbances if the error is zero to all disturbance polynomials of degree less than k but nonzero for a polynomial of degree k.

12 Black was 29 years old at the time. 13According to the story, many of Black’s colleagues at the Bell laboratories did not believe it was possible to feed back a signal 100 times as large as was the input and still keep the system stable. As will be discussed in Chapter 6, this dilemma was solved by H. Nyquist, also at the Labs.

206 Chapter 4 A First Analysis of Feedback •

•

Increasing the proportional feedback gain reduces steady-state errors but high gain almost always destabilizes the system. Integral control provides robust reduction in steady-state errors, but also may make the system less stable. Derivative control increases damping and improves stability. These three kinds of control combined form the classical PID controller. The standard PID controller is described by the equations  kI U(s) = kp + + kD s E(s) or s   1 + TD s E(s) = D(s)E(s). U(s) = kp 1 + TI s 

• •

•

This latter form is ubiquitous in the process-control industry and describes the basic controller in many control systems. Useful guidelines for tuning PID controllers were presented in Tables 4.2 and 4.3. A difference equation describing a digital controller to be used to replace a given analog controller can be found by replacing s with T2s z−1 z+1 in the transfer function and using z as a forward shift operator. Thus, if U(z) corresponds to u(kTs ) then zU(z) corresponds to u(kTs + Ts ). MATLAB can compute a discrete equivalent with the command c2d.

REVIEW QUESTIONS 1.

Give three advantages of feedback in control.

2.

Give two disadvantages of feedback in control.

3. A temperature control system is found to have zero error to a constant tracking input and an error of 0.5◦ C to a tracking input that is linear in time, rising at the rate of 40◦ C/ sec. What is the system type of this control system and what is the relevant error constant (Kp or Kv or etc.)? 4. What are the units of Kp , Kv , and Ka ? 5. What is the definition of system type with respect to reference inputs? 6. What is the definition of system type with respect to disturbance inputs? 7. Why does system type depend on where the external signal enters the system? 8. What is the main objective of introducing integral control? 9. What is the major objective of adding derivative control? 10. Why might a designer wish to put the derivative term in the feedback rather than in the error path? 11. What is the advantage of having a “tuning rule” for PID controllers? 12.

Give two reasons to use a digital controller rather than an analog controller.

13.

Give two disadvantages to using a digital controller.

14.

Give the substitution in the discrete operator z for the Laplace operator s if the approximation to the integral in Eq. (4.98) is taken to be the rectangle of height e(kTs ) and base Ts .

Problems

207

PROBLEMS Problems for Section 4.1: The Basic Equations of Control 4.1

If S is the sensitivity of the unity feedback system to changes in the plant transfer function and T is the transfer function from reference to output, show that S + T = 1.

4.2

We define the sensitivity of a transfer function G to one of its parameters k as the ratio of percent change in G to percent change in k. d ln G k dG dG/G = = . dk/k d ln k G dk

SkG =

The purpose of this problem is to examine the effect of feedback on sensitivity. In particular, we would like to compare the topologies shown in Fig. 4.23 for connecting three amplifier stages with a gain of −K into a single amplifier with a gain of −10. (a) For each topology in Fig. 4.23, compute βi so that if K = 10, Y = −10R. (b) For each topology, compute SkG when G = Y/R. [Use the respective βi values found in part (a).] Which case is the least sensitive? (c) Compute the sensitivities of the systems in Fig. 4.23(b,c) to β2 and β3 . Using your results, comment on the relative need for precision in sensors and actuators.

Figure 4.23 Three-amplifier topologies for Problem 4.2

b1

R

K

K

K

Y

(a)

R









K



K





b2






K

Y


b2

b2

(b)

R






K

K

K

Y


b3 (c)

4.3

Compare the two structures shown in Fig. 4.24 with respect to sensitivity to changes in the overall gain due to changes in the amplifier gain. Use the relation S=

K dF d ln F = d ln K F dK

as the measure. Select H1 and H2 so that the nominal system outputs satisfy F1 = F2 , and assume KH1 > 0.

208 Chapter 4 A First Analysis of Feedback R






K






K





F1

R






K

K

F2

 H1

H1

H2

(a)

(b)

Figure 4.24 Block diagrams for Problem 4.3

4.4

A unity feedback control system has the open-loop transfer function G(s) =

A . s(s + a)

(a) Compute the sensitivity of the closed-loop transfer function to changes in the parameter A. (b) Compute the sensitivity of the closed-loop transfer function to changes in the parameter a. (c) If the unity gain in the feedback changes to a value of β = 1, compute the sensitivity of the closed-loop transfer function with respect to β. 4.5

Compute the equation for the system error for the filtered feedback system shown in Fig. 4.4.

4.6

If S is the sensitivity of the filtered feedback system to changes in the plant transfer function and T is the transfer function from reference to output, compute the sum of S + T . Show that S + T = 1 if F = H. (a) Compute the sensitivity of the filtered feedback system respect to changes in the plant transfer function, G. (b) Compute the sensitivity of the filtered feedback system respect to changes in the controller transfer function, Dcl. (c) Compute the sensitivity of the filtered feedback system respect to changes in the filter transfer function, F. (d) Compute the sensitivity of the filtered feedback system respect to changes in the sensor transfer function, H.

shown in Fig. 4.4 with shown in Fig. 4.4 with shown in Fig. 4.4 with shown in Fig. 4.4 with

Problems for Section 4.2: Control of Steady-State Error 4.7

Consider the DC-motor control system with rate (tachometer) feedback shown in Fig. 4.25(a). (a) Find values for K
and kt
so that the system of Fig. 4.25(b) has the same transfer function as the system of Fig. 4.25(a). (b) Determine the system type with respect to tracking θr and compute the system Kv in terms of parameters K
and kt
. (c) Does the addition of tachometer feedback with positive kt increase or decrease Kv ?

4.8

Consider the system shown in Fig. 4.26, where (s + α)2 . D(s) = K 2 s + ωo2

Problems ur






Kp








Km s(1
tms)

K



1 k

u

ur








kts

K s(1
tms)

209 u

1
kts

(a)

(b)

Figure 4.25 Control system for Problem 4.7 Figure 4.26 Control system for Problem 4.8

R






1 s(s
1)

D(s)



Y

(a) Prove that if the system is stable, it is capable of tracking a sinusoidal reference input r = sin ωo t with zero steady-state error. (Look at the transfer function from R to E and consider the gain at ωo .) (b) Use Routh’s criterion to find the range of K such that the closed-loop system remains stable if ωo = 1 and α = 0.25. 4.9

Consider the system shown in Fig. 4.27, which represents control of the angle of a pendulum that has no damping. (a) What condition must D(s) satisfy so that the system can track a ramp reference input with constant steady-state error? (b) For a transfer function D(s) that stabilizes the system and satisfies the condition in part (a), find the class of disturbances w(t) that the system can reject with zero steady-state error.

Figure 4.27 Control system for Problem 4.9

W

R








D(s)
















1 s2

Y

K

4.10 A unity feedback system has the overall transfer function ωn2 Y (s) = T (s) = 2 . R(s) s + 2ζ ωn s + ωn2 Give the system type and corresponding error constant for tracking polynomial reference inputs in terms of ζ and ωn .

210 Chapter 4 A First Analysis of Feedback 4.11 Consider the second-order system 1 G(s) = 2 . s + 2ζ s + 1 We would like to add a transfer function of the form D(s) = K(s+a) (s+b) in series with G(s) in a unity feedback structure. (a) Ignoring stability for the moment, what are the constraints on K, a, and b so that the system is Type 1? (b) What are the constraints placed on K, a, and b so that the system is both stable and Type 1? (c) What are the constraints on a and b so that the system is both Type 1 and remains stable for every positive value for K? 4.12 Consider the system shown in Fig. 4.28(a). (a) What is the system type? Compute the steady-state tracking error due to a ramp input r(t) = ro t1(t). (b) For the modified system with a feed-forward path shown in Fig. 4.28(b), give the value of Hf so the system is Type 2 for reference inputs and compute the Ka in this case. (c) Is the resulting Type 2 property of this system robust with respect to changes in Hf ? i.e., will the system remain Type 2 if Hf changes slightly?

Figure 4.28 Control system for Problem 4.12

R




A s( s
1)

 

Y

(a)

Hf s R

Hr
















A s( s
1)

Y

(b)

4.13 A controller for a satellite attitude control with transfer function G = 1/s2 has been designed with a unity feedback structure and has the transfer function D(s) = 10(s+2) s+5 . (a) Find the system type for reference tracking and the corresponding error constant for this system. (b) If a disturbance torque adds to the control so that the input to the process is u + w, what is the system type and corresponding error constant with respect to disturbance rejection? 4.14 A compensated motor position control system is shown in Fig. 4.29. Assume that the sensor dynamics are H(s) = 1.

Problems Figure 4.29 Control system for Problem 4.14

211

W Compensator R







s
4 160 s
30





Plant






1 s(s
2)

Y

Sensor H(s)

(a) Can the system track a step reference input r with zero steady-state error? If yes, give the value of the velocity constant. (b) Can the system reject a step disturbance w with zero steady-state error? If yes, give the value of the velocity constant. (c) Compute the sensitivity of the closed-loop transfer function to changes in the plant pole at −2. (d) In some instances there are dynamics in the sensor. Repeat parts (a) to (c) for 20 and compare the corresponding velocity constants. H(s) = (s+20) 4.15 The general unity feedback system shown in Fig. 4.30 has disturbance inputs w1 , w2 , and w3 and is asymptotically stable. Also,  1  1 (s + z1i ) K1 m K2 m i=1 i=1 (s + z2i ) . G1 (s) = l m1 , G2 (s) = l m 1 1 2 s s i=1 (s + p1i ) i=1 (s + p2i ) (a) Show that the system is of Type 0, Type l1 , and Type (l1 + l2 ) with respect to disturbance inputs w1 , w2 , and w3 respectively.

Figure 4.30 Single input–single output unity feedback system with disturbance inputs

W1

R




W3

W2



G1(s)









G2(s)









Y



4.16 One possible representation of an automobile speed-control system with integral control is shown in Fig. 4.31.

Figure 4.31 System using integral control

W

Vc

k1








k1E

k2 s

F





 

m s k3

k1

V

212 Chapter 4 A First Analysis of Feedback (a) With a zero reference velocity input (vc = 0), find the transfer function relating the output speed v to the wind disturbance w. (b) What is the steady-state response of v if w is a unit ramp function? (c) What type is this system in relation to reference inputs? What is the value of the corresponding error constant? (d) What is the type and corresponding error constant of this system in relation to tracking the disturbance w? 4.17 For the feedback system shown in Fig. 4.32, find the value of α that will make the system Type 1 for K = 5. Give the corresponding velocity constant. Show that the system is not robust by using this value of α and computing the tracking error e = r − y to a step reference for K = 4 and K = 6.

Figure 4.32 Control system for Problem 4.17




a

R

K s
2





Y

4.18 Suppose you are given the system depicted in Fig. 4.33(a), where the plant parameter a is subject to variations.

R








4

U

x


1 s
a






1 s

Y

R






E(t)

G(s)

Y



1  4







(a)

(b)

Figure 4.33 Control system for Problem 4.18

(a) Find G(s) so that the system shown in Fig. 4.33(b) has the same transfer function from r to y as the system in Fig. 4.33(a). (b) Assume that a = 1 is the nominal value of the plant parameter. What is the system type and the error constant in this case? (c) Now assume that a = 1 + δa, where δa is some perturbation to the plant parameter. What is the system type and the error constant for the perturbed system? 4.19 Two feedback systems are shown in Fig. 4.34. (a) Determine values for K1 , K2 , and K3 so that (i) both systems exhibit zero steady-state error to step inputs (that is, both are Type 1), and (ii) their static velocity error constant Kv = 1 when K0 = 1.

Problems R








K1 s

U

K0 4s
1

Y

K2

R






U

K3



(a)

K0 4s
1

213

Y

(b)

Figure 4.34 Two feedback systems for Problem 4.19 (b) Suppose K0 undergoes a small perturbation: K0 → K0 + δK0 . What effect does this have on the system type in each case? Which system has a type which is robust? Which system do you think would be preferred? 4.20 You are given the system shown in Fig. 4.35, where the feedback gain β is subject to variations. You are to design a controller for this system so that the output y(t) accurately tracks the reference input r(t).

Figure 4.35 Control system for Problem 4.20

R






a



Di(s)

10 (s
1)(s
10)

Y

b

(a) Let β = 1. You are given the following three options for the controller Di (s): D1 (s) = kp ,

D2 (s) =

kp s + kI , s

D3 (s) =

kp s2 + kI s + k2 . s2

Choose the controller (including particular values for the controller constants) that will result in a Type 1 system with a steady-state error to a unit reference ramp of 1 . less than 10 (b) Next, suppose that there is some attenuation in the feedback path that is modeled by β = 0.9. Find the steady-state error due to a ramp input for your choice of Di (s) in part (a). (c) If β = 0.9, what is the system type for part (b)? What are the values of the appropriate error constant? 4.21 Consider the system shown in Fig. 4.36. (a) Find the transfer function from the reference input to the tracking error. (b) For this system to respond to inputs of the form r(t) = t n 1(t) (where n < q) with zero steady-state error, what constraint is placed on the open-loop poles p1 , p2 , . . . , pq ?

Figure 4.36 Control system for Problem 4.21

R








E

1 (s
p1)(s
p2) (s
pq)

Y

214 Chapter 4 A First Analysis of Feedback 4.22 A linear ODE model of the DC motor with negligible armature inductance (La = 0) and with a disturbance torque w was given earlier in the chapter; it is restated here, in slightly different form, as JRa Ra w, θ¨m + Ke θ˙m = υa + Kt Kt where θm is measured in radians. Dividing through by the coefficient of θ¨m , we obtain θ¨m + a1 θ˙m = b0 υa + c0 w, where

K t Ke Kt 1 , b0 = , c0 = . JRa JRa J With rotating potentiometers, it is possible to measure the positioning error between θ and the reference angle θr or e = θref − θm . With a tachometer we can measure the motor speed θ˙m . Consider using feedback of the error e and the motor speed θ˙m in the form υa = K(e − TD θ˙m ), a1 =

where K and TD are controller gains to be determined. (a) Draw a block diagram of the resulting feedback system showing both θm and θ˙m as variables in the diagram representing the motor. (b) Suppose the numbers work out so that a1 = 65, b0 = 200, and c0 = 10. If there is no load torque (w = 0), what speed (in rpm) results from va = 100 V? (c) Using the parameter values given in part (b), let the control be D = kp + kD s and find kp and kD so that, using the results of Chapter 3, a step change in θref with zero load torque results in a transient that has an approximately 17% overshoot and that settles to within 5% of steady-state in less than 0.05 sec. (d) Derive an expression for the steady-state error to a reference angle input, and compute its value for your design in part (c) assuming θref = 1 rad. (e) Derive an expression for the steady-state error to a constant disturbance torque when θref = 0 and compute its value for your design in part (c) assuming w = 1.0. 4.23 We wish to design an automatic speed control for an automobile. Assume that (1) the car has a mass m of 1000 kg, (2) the accelerator is the control U and supplies a force on the automobile of 10 N per degree of accelerator motion, and (3) air drag provides a friction force proportional to velocity of 10 N · sec/m. (a) Obtain the transfer function from control input U to the velocity of the automobile. (b) Assume the velocity changes are given by V (s) =

1 0.05 U(s) + W (s), s + 0.02 s + 0.02

where V is given in meters per second, U is in degrees, and W is the percent grade of the road. Design a proportional control law U = −kp V that will maintain a velocity error of less than 1 m/sec in the presence of a constant 2% grade. (c) Discuss what advantage (if any) integral control would have for this problem. (d) Assuming that pure integral control (that is, no proportional term) is advantageous, select the feedback gain so that the roots have critical damping (ζ = 1). 4.24 Consider the automobile speed control system depicted in Fig. 4.37.

Problems Figure 4.37 Automobile speed-control system

215

W R  Desired speed Y  Actual speed W  Road grade, %




Hr

R



B s
a

A s
a

kp









Y

Hy

(a) Find the transfer functions from W (s) and from R(s) to Y (s). (b) Assume that the desired speed is a constant reference r, so that R(s) = ro/s. Assume that the road is level, so w(t) = 0. Compute values of the gains K, Hr , and Hf to guarantee that lim y(t) = ro . t→∞

Include both the open-loop (assuming Hy = 0) and feedback cases (Hy = 0) in your discussion. (c) Repeat part (b) assuming that a constant grade disturbance W (s) = wo/s is present in addition to the reference input. In particular, find the variation in speed due to the grade change for both the feed-forward and feedback cases. Use your results to explain (1) why feedback control is necessary and (2) how the gain kp should be chosen to reduce steady-state error. (d) Assume that w(t) = 0 and that the gain A undergoes the perturbation A + δA. Determine the error in speed due to the gain change for both the feed-forward and feedback cases. How should the gains be chosen in this case to reduce the effects of δA? 4.25 Consider the multivariable system shown in Fig. 4.38. Assume that the system is stable. Find the transfer functions from each disturbance input to each output and determine the steady-state values of y1 and y2 for constant disturbances. We define a multivariable system to be type k with respect to polynomial inputs at wi if the steady-state value of every output is zero for any combination of inputs of degree less than k and at least one input is a non zero constant for an input of degree k. What is the system type with respect to disturbance rejection at w1 ? At w2 ?

Figure 4.38 Multivariable system

W1

R1




1 s



 R2








1 s
1




W2




1 s
1









1 s
2









Y1

Y2

216 Chapter 4 A First Analysis of Feedback Problems for Section 4.3: The Three-Term Controller. PID Control 4.26 The transfer functions of speed control for a magnetic tape-drive system are shown in Fig. 4.39. The speed sensor is fast enough that its dynamics can be neglected and the diagram shows the equivalent unity feedback system. (a) Assuming the reference is zero, what is the steady-state error due to a step disturbance torque of 1 N · m? What must the amplifier gain K be in order to make the steady-state error ess ≤ 0.01 rad/sec? (b) Plot the roots of the closed-loop system in the complex plane, and accurately sketch the time response of the output for a step reference input using the gain K computed in part (a). (c) Plot the region in the complex plane of acceptable closed-loop poles corresponding to the specifications of a 1% settling time of ts ≤ 0.1 sec and an overshoot Mp ≤ 5%. (d) Give values for kp and kD for a PD controller, which will meet the specifications. (e) How would the disturbance-induced steady-state error change with the new control scheme in part (d)? How could the steady-state error to a disturbance torque be eliminated entirely?

Figure 4.39 Speed-control system for a magnetic tape-drive

Disturbance torque Amplifier


Reference speed, r



Torque motor 10 0.5s
1

K



Torque


Tape dynamics




1 Js
b



m

J  0.10 kg m2 b  1.00 N m sec

4.27 Consider the system shown in Fig. 4.40 with PI control. (a) (b) (c) (d)

Figure 4.40 Control system for Problem 4.27

Determine the transfer function from R to Y . Determine the transfer function from W to Y . What is the system type and error constant with respect to reference tracking? What is the system type and error constant with respect to disturbance rejection? W

R




 

kP s
k I s

U







10 s2
s
20

4.28 Consider the second-order plant with transfer function G(s) = and in a unity feedback structure.

1 (s + 1)(5s + 1)

Y

Problems

217

(a) Determine the system type and error constant with respect to tracking polynomial reference inputs of the system for P [D = kp ], PD [D = kp + kD s], and PID [D = kp + kI/s + kD s] controllers. Let kp = 19, kI = 0.5, and kD = 4/19. (b) Determine the system type and error constant of the system with respect to disturbance inputs for each of the three regulators in part (a) with respect to rejecting polynomial disturbances w(t) at the input to the plant. (c) Is this system better at tracking references or rejecting disturbances? Explain your response briefly. (d) Verify your results for parts (a) and (b) using MATLAB by plotting unit step and ramp responses for both tracking and disturbance rejection. 4.29 The DC-motor speed control shown in Fig. 4.41 is described by the differential equation y˙ + 60y = 600va − 1500w, where y is the motor speed, va is the armature voltage, and w is the load torque. Assume the armature voltage is computed using the PI control law  

t va = − kp e + kI edt , 0

where e = r − y. (a) Compute the transfer function from W to Y as a function of kp and kI . (b) Compute values for kp and kI so that the characteristic equation of the closed-loop system will have roots at −60 ± 60j.

Figure 4.41 DC Motor speed-control block diagram for Problems 4.29 and 4.30

W

1500

R






e

D

ya

 600








1 s
60

Y

4.30 For the system in Problem 4.29, compute the following steady-state errors: (a) (b) (c) (d) (e)

to a unit-step reference input; to a unit-ramp reference input; to a unit-step disturbance input; for a unit-ramp disturbance input. Verify your answers to (a) and (d) using MATLAB. Note that a ramp response can be generated as a step response of a system modified by an added integrator at the reference input.

4.31 Consider the satellite-attitude control problem shown in Fig. 4.42 where the normalized parameters are J = 10

spacecraft inertia, N·m·sec2 /rad

θr = reference satellite attitude, rad.

218 Chapter 4 A First Analysis of Feedback Figure 4.42 Satellite attitude control

w ur

Hr






D(s)











1 Js

1 s

u

Hy

θ = actual satellite attitude, rad. Hy = 1

sensor scale, factor V/rad.

Hr = 1

reference sensor scale factor, V/rad.

w = disturbance torque. N·m (a) Use proportional control, P, with D(s) = kp , and give the range of values for kp for which the system will be stable. (b) Use PD control and let D(s) = (kp + kD s) and determine the system type and error constant with respect to reference inputs. (c) Use PD control, let D(s) = (kp + kD s) and determine the system type and error constant with respect to disturbance inputs. (d) Use PI control, let D(s) = (kp + kI/s), and determine the system type and error constant with respect to reference inputs. (e) Use PI control, let D(s) = (kp + kI/s), and determine the system type and error constant with respect to disturbance inputs. (f) Use PID control, let D(s) = (kp + kI/s + kD s) and determine the system type and error constant with respect to reference inputs. (g) Use PID control, let D(s) = (kp + kI/s + kD s) and determine the system type and error constant with respect to disturbance inputs. 4.32 The unit-step response of a paper machine is shown in Fig. 4.43(a) where the input into the system is stock flow onto the wire and the output is basis weight (thickness). The time delay and slope of the transient response may be determined from the figure. (a) Find the proportional, PI, and PID-controller parameters using the Zeigler–Nichols transient-response method. (b) Using proportional feedback control, control designers have obtained a closed-loop system with the unit impulse response shown in Fig. 4.43(b). When the gain Ku = 8.556, the system is on the verge of instability. Determine the proportional-, PI-, and PID-controller parameters according to the Zeigler–Nichols ultimate sensitivity method. 4.33 A paper machine has the transfer function G(s) =

e−2s , 3s + 1

where the input is stock flow onto the wire and the output is basis weight or thickness. (a) Find the PID-controller parameters using the Zeigler–Nichols tuning rules. (b) The system becomes marginally stable for a proportional gain of Ku = 3.044 as shown by the unit impulse response in Fig. 4.44. Find the optimal PID-controller parameters according to the Zeigler–Nichols tuning rules.

Problems 1.0 Unit impulse response

0.4

Step response

0.8 0.6 0.4 0.2 0.0

219

0

1

2

3

4 5 6 7 Time (sec)

8

0.3 0.2 0.1 0.0 0.1 0.2 0.3 0.4

9 10

0

1

2

3

(a)

4 5 6 7 Time (sec)

8

9 10

(b)

Figure 4.43 Paper-machine response data for Problem 4.32 Figure 4.44 Unit impulse response for the paper machine in Problem 4.33

0.020 0.015 0.010 y

0.005 0.00 0.005 0.010 0.015 0.0



5.0

10.0 15.0 Time (sec)

20.0

25.0

Problems for Section 4.4: Introduction to Digital Control 4.34 Compute the discrete equivalents for the following possible controllers using the trapezoid rule of Eq. (4.104). Let Ts = 0.05 in each case. (a) D1 (s) = (s + 2)/2 s+2 (b) D2 (s) = 2 s+4 (s + 2) (c) D3 (s) = 5 s + 10 (s + 2)(s + 0.1) (d) D4 (s) = 5 (s + 10)(s + 0.01) 4.35 Give the difference equations corresponding to the discrete controllers found in Problem 4.34 respectively. (a) (b) (c) (d)

part 1 part 2 part 3 part 4

5

The Root-Locus Design Method

A Perspective on the Root-Locus Design Method In Chapter 3 we related the features of a step response, such as rise time, overshoot, and settling time, to pole locations in the s-plane of the transform of a second-order system characterized by the natural frequency ωn , the damping ratio ζ , and the real part σ . This relationship is shown graphically in Fig. 3.15. We also examined the changes in these transient-response features when a pole or a zero is added to the transfer function. In Chapter 4 we saw how feedback can improve steady-state errors and can also influence dynamic response by changing the system’s pole locations. In this chapter we present a specific technique that shows how changes in one of a system’s parameters will modify the roots of the characteristic equation, which are the closedloop poles, and thus change the system’s dynamic response. The method was developed by W. R. Evans who gave rules for plotting the paths of the roots, a plot he called the Root Locus. With the development of MATLAB
and similar software the rules are no longer needed for detailed plotting, but we feel it is essential for a control designer to understand how proposed dynamic controllers will influence a locus as a guide in the design process. We also feel that it is important to understand the basics of how loci are generated in order to perform sanity checks on the computer results. For these reasons, study of the Evans rules is important. The root locus is most commonly used to study the effect of loop gain variations; however, the method is general and can be used to plot the roots of

220

5.1 Root Locus of a Basic Feedback System

221

any polynomial with respect to any one real parameter that enters the equation linearly. For example, the root-locus method can be used to plot the roots of a characteristic equation as the gain of a velocity sensor feedback changes, or the parameter can be a physical parameter such as motor inertia or armature inductance. Finally, a root locus can be plotted for a characteristic equation that results from the analysis of digital control systems using the z-transform, a topic we introduced in Chapter 4 and will discuss further in Chapter 8.

Chapter Overview We open in Section 5.1 by illustrating the root locus for some simple feedback systems for which the equations can be solved directly. In Section 5.2 we show how to put an equation into the proper form for developing the rules for the root-locus behavior. In Section 5.3 this approach is applied to determine the locus for a number of typical control problems, which illustrate the factors that influence the final shape. MATLAB is used for detailed plotting of specific loci. When adjustment of the selected parameter alone cannot produce a satisfactory design, designs using other parameters can be studied or dynamic elements such as lead, lag, or notch compensations can be introduced, as described in Section 5.4. In Section 5.5 the uses of the root locus for design are summarized by a comprehensive design for the attitude control of a small airplane. In Section 5.6, the root-locus method is extended to guide the design of systems with a negative parameter, systems with more than one variable parameter, and systems with simple time delay. Finally, Section 5.7 gives historical notes on the origin of root-locus design.

5.1 Root Locus of a Basic Feedback System We begin with the basic feedback system shown in Fig. 5.1. For this system, the closed-loop transfer function is Y (s) D(s)G(s) = T (s) = , R(s) 1 + D(s)G(s)H(s)

(5.1)

and the characteristic equation, whose roots are the poles of this transfer function, is 1 + D(s)G(s)H(s) = 0. Figure 5.1 Basic closed-loop block diagram

(5.2)

W

R




Controller





D(s)

U





Plant



Y

G(s)


 Sensor H(s)




V

222 Chapter 5 The Root-Locus Design Method To put the equation in a form suitable for study of the roots as a parameter changes, we first put the equation in polynomial form and select the parameter of interest, which we will call K. We assume that we can define component polynomials a(s) and b(s) so that the characteristic polynomial is in the form a(s) + Kb(s). We then define the b(s) transfer function L(s) = a(s) so that the characteristic equation can be written as1 1 + KL(s) = 0.

Evans’s method

(5.3)

If, as is often the case, the parameter is the gain of the controller, then L(s) is simply proportional to D(s)G(s)H(s). Evans suggested that we plot the locus of all possible roots of Eq. (5.3) as K varies from zero to infinity and then use the resulting plot to aid us in selecting the best value of K. Furthermore, by studying the effects of additional poles and zeros on this graph, we can determine the consequences of additional dynamics added to D(s) as compensation in the loop. We thus have a tool not only for selecting the specific parameter value but for designing the dynamic compensation as well. The graph of all possible roots of Eq. (5.3) relative to parameter K is called the root locus, and the set of rules to construct this graph is called the root-locus method of Evans. We begin our discussion of the method with the mechanics of constructing a root locus, using the equation in the form of Eq. (5.3) and K as the variable parameter. To set the notation for our study, we assume here that the transfer function L(s) is a rational function whose numerator is a monic2 polynomial b(s) of degree m and whose denominator is a monic polynomial a(s) of degree n such that3 n ≥ m. We can factor these polynomials as b(s) = sm + b1 sm−1 + · · · + bm = (s − z1 )(s − z2 ) · · · (s − zm ) =

m 

(s − zi ),

(5.4)

i=1

a(s) = sn + a1 sn−1 + · · · + an =

n 

(s − pi ).

i=1

The roots of b(s) = 0 are the zeros of L(s) and are labeled zi , and the roots of a(s) = 0 are the poles of L(s) and are labeled pi . The roots of the characteristic equation itself are ri from the factored form (n > m), a(s) + Kb(s) = (s − r1 )(s − r2 ) · · · (s − rn ).

(5.5)

1 In the most common case, L(s) is the loop transfer function of the feedback system and K is the gain of

the controller–plant combination. However, the root locus is a general method suitable for the study of any polynomial and any parameter that can be put in the form of Eq. (5.3). 2 Monic means that the coefficient of the highest power of s is 1. 3 If L(s) is the transfer function of a physical system, it is necessary that n ≥ m or else the system would

have an infinite response to a finite input. If the parameter should be chosen so that n < m, then we can consider the equivalent equation 1 + K −1 L(s)−1 = 0.

5.1 Root Locus of a Basic Feedback System

223

We may now state the root-locus problem expressed in Eq. (5.3) in several equivalent but useful ways. Each of the following equations has the same roots: 1 + KL(s) = 0, b(s) 1+K = 0, a(s) a(s) + Kb(s) = 0,

Root-locus forms

EXAMPLE 5.1

(5.6) (5.7) (5.8)

1 L(s) = − . (5.9) K Equations (5.6)–(5.9) are sometimes referred to as the root-locus form or Evans form of a characteristic equation. The root locus is the set of values of s for which Eqs. (5.6)–(5.9) hold for some positive real value4 of K. Because the solutions to Eqs. (5.6)–(5.9) are the roots of the closed-loop system characteristic equation and are thus closed-loop poles of the system, the root-locus method can be thought of as a method for inferring dynamic properties of the closed-loop system as the parameter K changes.

Root Locus of a Motor Position Control In Chapter 2 we saw that a normalized transfer function of a DC motor voltage-toposition can be Y (s) A m (s) = = G(s) = . Va (s) U(s) s(s + c) Solve for the root locus of closed-loop poles of the system created by feeding back the output m as shown in Fig. 5.1 with respect to the parameter A if D(s) = H(s) = 1 and also c = 1. Solution. In terms of our notation, the values are 1 , b(s) = 1, m = 0, zi = {empty}, (5.10) L(s) = s(s + 1) K = A,

a(s) = s2 + s,

n = 2,

pi = 0, −1.

From Eq. (5.6) the root locus is a graph of the roots of the quadratic equation a(s) + Kb(s) = s2 + s + K = 0.

(5.11)

Using the quadratic formula, we can immediately express the roots of Eq. (5.11) as √ 1 − 4K 1 . (5.12) r1 , r2 = − ± 2 2 A plot of the corresponding root locus is shown in Fig. 5.2. For 0 ≤ K ≤ 1/4, the roots are real between −1 and 0. At K = 1/4 there are two roots at − 1/2, and for K > 1/4 the roots become complex with real parts constant at − 1/2 and imaginary parts that increase essentially in proportion to the square root of K. The dashed lines in Fig. 5.2 correspond to roots with a damping ratio ζ = 0.5. The poles of L(s) at s = 0 and 4 If K is positive, the locus is called the “positive” locus. We will consider later the simple changes if K < 0, resulting in a “negative” locus.

224 Chapter 5 The Root-Locus Design Method 1.5 u  sin1 z30

1 Imaginary axis

Figure 5.2 Root locus for 1 L(s) = s(s+1)

0.5 0 0.5 1 1.5 2

1

0 Real axis

1

2

s = −1 are marked by the symbol ×, and the points where the locus crosses the lines where the damping ratio equals 0.5 are marked with dots (•). We can compute K at the point where the locus crosses ζ = 0.5 because we know that, √if ζ = 0.5, then θ = 30◦ and the magnitude of the imaginary part of the root is 3 times the magnitude of the real part. Since the size of the real part is 21 , from Eq. (5.12) we have √ √ 4K − 1 3 = , 2 2 and, therefore, K = 1.

Breakaway points are where roots move away from the real axis

EXAMPLE 5.2

We can observe several features of this simple locus by looking at Eqs. (5.11) and (5.12) and Fig. 5.2. First, there are two roots and thus two branches of the root locus. At K = 0 these branches begin at the poles of L(s) (which are at 0 and −1), as they should, since for K = 0 the system is open loop and the characteristic equation is a(s) = 0. As K is increased, the roots move toward each other, coming together at s = − 21 , and at that point they break away from the real axis. After the breakaway point the roots move off to infinity with equal real parts, so the sum of the two roots is always −1. From the viewpoint of design, we see that by altering the value of the parameter K, we can cause the closed-loop poles to be at any point along the locus in Fig. 5.2. If some points along this locus correspond to a satisfactory transient response, then we can complete the design by choosing the corresponding value of K; otherwise, we are forced to consider a more complex controller. As we pointed out earlier, the root locus technique is not limited to focusing on the system gain (K = A in Example 5.1); the same ideas are applicable for finding the locus with respect to any parameter that enters linearly in the characteristic equation.

Root Locus with Respect to a Plant Open-Loop Pole Consider the characteristic equation as in Example 5.1, except that now let D(s) = H(s) = 1 and also let A = 1. Select c as the parameter of interest in the equation 1 . (5.13) 1 + G(s) = 1 + s(s + c) Find the root locus of the characteristic equation with respect to c.

5.1 Root Locus of a Basic Feedback System

225

Solution. The corresponding closed-loop characteristic equation in polynomial form is s2 + cs + 1 = 0. (5.14) The alternatives of Eq. (5.6) with the associated definitions of poles and zeros will apply if we let L= 2s , b(s) = s, m = 1, zi = 0, s +1 (5.15) n = 2, pi = +j, −j. K = c, a(s) = s2 + 1, Thus, the root-locus form of the characteristic equation is s 1+c 2 = 0. s +1 The solutions to Eq. (5.14) are easily computed as √ c2 − 4 c r1 , r2 = − ± . (5.16) 2 2 The locus of solutions is shown in Fig. 5.3, with the poles [roots of a(s)] again indicated by ×’s and the zero [root of b(s)] by the circle ( ) symbol. Note that when c = 0, the roots are at the ×’s on the imaginary axis and the corresponding response would be oscillatory. The damping ratio ζ grows as c increases from 0. At c = 2, there are two roots at s = −1, and the two locus segments abruptly change direction and move in opposite directions along the real axis; this point of multiple roots where two or more roots come into the real axis is called a break-in point.

◦

Break-in point

Of course, computing the root locus for a quadratic equation is easy to do since we can solve the characteristic equation for the roots, as was done in Eqs. (5.12) and (5.16), and directly plot these as a function of the parameter K. To be useful, the method must be suitable for higher-order systems for which explicit solutions are difficult to obtain and rules for the construction of a general root locus were developed by Evans. With the availability of MATLAB, these rules are no longer necessary to plot a specific locus. The command rlocus(sys) will do that. However, in control design we are interested not only in a specific locus but also in how to modify the dynamics in 1.5 1 Imaginary axis

Figure 5.3 Root locus vs. damping factor c for 1 + G(s) = 1 1 + s(s+c) =0

0.5 䊊

0 0.5 1 1.5 2

1

0 Real axis

1

2

226 Chapter 5 The Root-Locus Design Method such a way as to propose a system that will meet the dynamic response specifications for good control performance. For this purpose, it is very useful to be able to roughly sketch a locus so as to be able to evaluate the consequences of possible compensation alternatives. It is also important to be able to quickly evaluate the correctness of a computer-generated locus to verify that what is plotted by MATLAB is in fact what was meant to be plotted. It is easy to get a constant wrong or to leave out a term and GIGO5 is the well-known first rule of computation.

5.2 Guidelines for Determining a Root Locus We begin with a formal definition of a root locus. From the form of Eq. (5.6), we define the root locus this way: Definition I. The root locus is the set of values of s for which 1+KL(s) = 0 is satisfied as the real parameter K varies from 0 to +∞. Typically, 1+KL(s) = 0 is the characteristic equation of the system, and in this case the roots on the locus are the closed-loop poles of that system. Now suppose we look at Eq. (5.9). If K is to be real and positive, L(s) must be real and negative. In other words, if we arrange L(s) in polar form as magnitude and phase, then the phase of L(s) must be 180◦ in order to satisfy Eq. (5.9). We can thus define the root locus in terms of this phase condition as follows. The basic root-locus rule; the phase of L(s) = 180◦

Definition II. The root locus of L(s) is the set of points in the s-plane where the phase of L(s) is 180◦ . If we define the angle to the test point from a zero as ψi and the angle to the test point from a pole as φi then Definition II is expressed as those points in the s-plane where, for integer l, 

ψi −



φi = 180◦ + 360◦ (l − 1).

(5.17)

The immense merit of Definition II is that, while it is very difficult to solve a highorder polynomial by hand, computing the phase of a transfer function is relatively easy. The usual case is when K is real and positive, and we call this case the positive or 180◦ locus. When K is real and negative, L(s) must be real and positive with a phase of 0◦ , and this case is called the negative or 0◦ locus. From Definition II we can, in principle, determine a positive root locus for a complex transfer function by measuring the phase and marking those places where we find 180◦ . This direct approach can be illustrated by considering the example L(s) =

s+1 . s(s + 5)[(s + 2)2 + 4]

(5.18)

◦

In Fig. 5.4 the poles of this L(s) are marked × and the zero is marked . Suppose we select the test point s0 = −1 + 2j. We would like to test whether or not s0 lies on 5 Garbage in, Garbage out.

5.2 Guidelines for Determining a Root Locus Figure 5.4 Measuring the phase of Eq. (5.18)

227

3 f2  0

Imaginary axis

2

s0

1

1

䊊

c1

f3

2 3 6

f1

f4

0

4

2 Real axis

0

2

the root locus for some value of K. For this point to be on the locus, we must have ∠L(s0 ) = 180◦ + 360◦ (l − 1) for some integer l, or equivalently, from Eq. (5.18), ∠(s0 + 1) − ∠s0 − ∠(s0 + 5) − ∠[(s0 + 2)2 + 4] = 180◦ + 360◦ (l − 1). (5.19) The angle from the zero term s0 + 1 can be computed6 by drawing a line from the location of the zero at −1 to the test point s0 . In this case the line is vertical and has a phase angle marked ψ1 = 90◦ in Fig. 5.4. In a similar fashion, the vector from the pole at s = 0 to the test point s0 is shown with angle φ1 , and the angles of the two vectors from the complex poles at −2 ± 2j to s0 are shown with angles φ2 and φ3 . The phase of the vector s0 + 5 is shown with angle φ4 . From Eq. (5.19) we find the total phase of L(s) at s = s0 to be the sum of the phases of the numerator term corresponding to the zero minus the phases of the denominator terms corresponding to the poles: ∠L = ψ1 − φ1 − φ2 − φ3 − φ4 = 90◦ − 116.6◦ − 0◦ − 76◦ − 26.6◦ = −129.2◦ . Since the phase of L(s) is not 180◦ , we conclude that s0 is not on the root locus, so we must select another point and try again. Although measuring phase is not particularly hard, measuring phase at every point in the s-plane is hardly practical. Therefore, to make the method practical, we need some general guidelines for determining where the root locus is. Evans developed a set of rules for the purpose, which we will illustrate by applying them to the root locus for 1 L(s) = . (5.20) s[(s + 4)2 + 16] We begin by considering the positive locus, which is by far the most common case.7 The first three rules are relatively simple to remember and are essential for any reasonable sketch. The last two are less useful but are used occasionally. As usual, we assume that MATLAB or its equivalent is always available to make an accurate plot of a promising locus. 6 The graphical evaluation of the magnitude and phase of a complex number is reviewed in Appendix WD,

Section 3. 7 The negative locus will be considered in Section 5.6.

228 Chapter 5 The Root-Locus Design Method 5.2.1 Rules for Plotting a Positive (180◦ ) Root Locus RULE 1. The n branches of the locus start at the poles of L(s) and m of these branches end on the zeros of L(s). From the equation a(s) + Kb(s) = 0, if K = 0, the equation reduces to a(s) = 0, whose roots are the poles. When K approaches infinity, s must be such that either b(s) = 0 or s → ∞. Since there are m zeros where b(s) = 0, m branches can end in these places. The case for s → ∞ is considered in Rule 3. RULE 2. The loci are on the real axis to the left of an odd number of poles and zeros. If we take a test point on the real axis, such as s0 in Fig. 5.5, we find that the angles φ1 and φ2 of the two complex poles cancel each other, as would the angles from complex conjugate zeros. Angles from real poles or zeros are 0◦ if the test point is to the right and 180◦ if the test point is to the left of a given pole or zero. Therefore, for the total angle to add to 180◦ + 360◦ (l − 1), the test point must be to the left of an odd number of real-axis poles plus zeros as shown in Fig. 5.5. RULE 3. For large s and K, n − m of the loci are asymptotic to lines at angles φl radiating out from the point s = α on the real axis, where 180◦ + 360◦ (l − 1) φl = , l = 1, 2, . . . , n − m, (5.21) n−m   pi − zi α= . n−m As K → ∞, the equation 1 L(s) = − K can be satisfied only if L(s) = 0. This can occur in two apparently different ways. In the first instance, as discussed in Rule 1, m roots will be found to approach the zeros of L(s). The second manner in which L(s) may go to zero is if s → ∞ since, by assumption, n is larger than m. The asymptotes describe how these n − m roots approach s → ∞. For large s, the equation 1+K 4

f2

2 Imaginary axis

Figure 5.5 Rule 2. The real-axis parts of the locus are to the left of an odd number of poles and zeros

sm + b1 sm−1 + · · · + bm =0 sn + a1 sn−1 + · · · + an

f3  0

s0

2 4 8

f1 6

4

2 Real axis

0

1

2

(5.22)

5.2 Guidelines for Determining a Root Locus

229

can be approximated8 by 1 = 0. (5.23) (s − α)n−m This is the equation for a system in which there are n − m poles, all clustered at s = α. Another way to visualize this same result is to consider the picture we would see if we could observe the locations of poles and zeros from a vantage point of very large s: They would appear to cluster near the s-plane origin. Thus, m zeros would cancel the effects of m of the poles, and the other n − m poles would appear to be in the same place. We say that the locus of Eq. (5.22) is asymptotic to the locus of Eq. (5.23) for large values of K and s. We need to compute α and to find the locus for the resulting asymptotic system. To find the locus, we choose our search point s0 such that s0 = Rejφ for some large fixed value of R and variable φ. Since all poles of this simple system are in the same place, the angle of its transfer function is 180◦ if all n − m angles, each equal to φl , sum to 180◦ . Therefore, φl is given by 1+K

(n − m)φl = 180◦ + 360◦ (l − 1) The angles of the asymptotes

for some integer l. Thus, the asymptotic root locus consists of radial lines at the n − m distinct angles given by 180◦ + 360◦ (l − 1) , l = 1, 2, . . . , n − m. (5.24) φl = n−m For the system described by Eq. (5.20), n − m = 3 and φ1,2,3 = 60◦ , 180◦ , and 300◦ or ±60◦ , 180◦ . The lines of the asymptotic locus come from s0 = α on the real axis. To determine α, we make use of a simple property of polynomials. Suppose we consider the monic polynomial a(s) with coefficients ai and roots pi , as in Eq. (5.4), and we equate the polynomial form with the factored form sn + a1 sn−1 + a2 sn−2 + · · · + an = (s − p1 )(s − p2 ) · · · (s − pn ). If we multiply out the factors on the right side of this equation, we see that the · · · − pn . On the left side of the equation, we see coefficient of sn−1 is −p1 − p2 − that this term is a1 . Thus a1 = − pi ; in other words, the coefficient of the second highest term in a monic polynomial is the negative sum of its roots—in this case, the poles of L(s). Applying this result to the polynomial b(s), we find the negative sum of the zeros to be b1 . These results can be written as  −b1 =  zi , (5.25) −a1 = pi . Finally, we apply this result to the closed-loop characteristic polynomial obtained from Eq. (5.22): sn + a1 sn−1 + · · · + an + K(sm + b1 sm−1 + · · · + bm )

(5.26)

= (s − r1 )(s − r2 ) · · · (s − rn ) = 0. Note that the sum of the roots is the negative of the coefficient of sn−1 and is independent of K if m < n − 1. Therefore, if L(s) has at least two more poles than zeros, 8 This approximation can be obtained by dividing a(s) by b(s) and matching the dominant two terms (highest powers in s) to the expansion of (s − α)n−m .

230 Chapter 5 The Root-Locus Design Method  we have a1 = − ri . We have thus shown that the center point of the roots does not change with K if m < n − 1 and that the open-loop and closed-loop sum is the same and is equal to −a1 , which can be expressed as   pi . (5.27) − ri = − For large values of K, we have seen that m of the roots ri approach the zeros zi and 1 n − m of the roots approach the branches of the asymptotic system (s−α) n−m whose poles add up to (n − m)α. Combining these results we conclude that the sum of all the roots equals the sum of those roots that go to infinity plus the sum of those roots that go to the zeros of L(s):    zi = − pi . − ri = −(n − m)α − The center of the asymptotes

Solving for α, we get

 pi − z i α= . (5.28) n−m   Notice that in the sums pi and zi the imaginary parts always add to zero, since complex poles and zeros always occur in complex conjugate pairs. Thus Eq. (5.28) requires information about the real parts only. For Eq. (5.20), −4 − 4 + 0 α= 3−0 8 = − = −2.67. 3 shown dashed in Fig. 5.6. Notice that they cross the The asymptotes at ±60◦ are √ imaginary axis at ±(2.67)j 3 = ±4.62j. The asymptote at 180◦ was already found on the real axis by Rule 2. 

RULE 4. The angle(s) of departure of a branch of the locus from a pole of multiplicity q is given by   ψi − φi − 180◦ − 360◦ (l − 1), (5.29) qφl,dep = i =l

and the angle(s) of arrival of a branch at a zero of multiplicity q is given by   φi − ψi + 180◦ + 360◦ (l − 1). qψl,arr = i =l

6 4 Imaginary axis

Figure 5.6 The asymptotes are n − m radial lines from α at equal angles

2 0 2 4 6 10

5

0 Real axis

5

(5.30)

5.2 Guidelines for Determining a Root Locus 6

s0

4 Imaginary axis

Figure 5.7 The departure and arrival angles are found by looking near a pole or zero

231

f1

Pole 2

2

f3

0

Pole 3

2

f2

4

Pole 1 6 10

5

0

5

Real axis

If a system has poles near the imaginary axis it can be important to know if the locus, which starts at such a pole, starts off toward the stable left half-plane (LHP) or heads toward the unstable right half-plane (RHP). To compute the angle by which a branch of the locus departs from one of the poles we take a test point so very near the pole in question, define the angle from that pole to the test point as φl,dep and transpose all other terms of Eq.(5.17) to the right-hand side. We can illustrate the process by taking the test point so to be near the pole at −4 + 4j of our example and computing the angle of L(s0 ). The situation is sketched in Fig. 5.7, and the angle from −4 + 4j to the test point we define as φ1 . We select the test point close enough to the pole that the angles φ2 and φ3 to the test point can be considered the same as those angles to the pole. Thus, φ2 = 90◦ , φ3 = 135◦ , and φ1 can be calculated from the angle condition as whatever it takes to make the total be 180◦ . The calculation is (l = 1) φ1 = −90◦ − 135◦ − 180◦

(5.31)

= −405◦

(5.32)

◦

= −45 .

Rule for departure angles

(5.33)

By the complex conjugate symmetry of the plots, the angle of departure of the locus near the pole at −4 − 4j will be +45◦ . If there had been zeros in L(s), the angles from the pole to the zeros would have been added to the right side of Eq. (5.31). For the general case, we can see from Eq. (5.31) that the angle of departure from a single pole is   φ1,dep = ψi − φi − 180◦ , (5.34) i =1

 where φi is the sum of the angles to the remaining poles and ψi is the sum of the angles to all the zeros. For a multiple pole of order q, we must count the angle from the pole q times. This alters Eq. (5.34) to   qφl,dep = ψi − φi − 180◦ − 360◦ (l − 1), (5.35) 

i =l

where l takes on q values because there are q branches of the locus that depart from such a multiple pole.

232 Chapter 5 The Root-Locus Design Method

Rule for arrival angles

The process of calculating a departure angle for small values of K, as shown in Fig. 5.7, is also valid for computing the angle by which a root locus arrives at a zero of L(s) for large values of K. The general formula that results is   qψl,arr = φi − ψi + 180◦ + 360◦ (l − 1) (5.36) i =l



where φi is the sum of the angles to all the poles, the remaining zeros, and l is an integer as before.



ψi is the sum of the angles to

RULE 5. The locus can have multiple roots at points on the locus and the branches will approach a point of q roots at angles separated by 180◦ + 360◦ (l − 1) q

Continuation locus

(5.37)

and will depart at angles with the same separation. As with any polynomial, it is possible for a characteristic polynomial of a degree greater than 1 to have multiple roots. For example, in the second-order locus of Fig. 5.2, there are two roots at s = − 1/2 when K = 1/4. Here the horizontal branches of the locus come together and the vertical branches break away from the real axis, becoming complex for K > 1/4. The locus arrives at 0◦ and 180◦ and departs at +90◦ and −90◦ . In order to compute the angles of arrival and departure from a point of multiple roots, it is useful to use a trick we call the continuation locus. We can imagine plotting a root locus for an initial range of K, perhaps for 0 ≤ K ≤ K1 . If we let K = K1 + K2 , we can then plot a new locus with parameter K2 , a locus which is the continuation of the original locus and whose starting poles are the roots of the original system at K = K1 . To see how this works, we return to the second-order root locus of Eq. (5.11) and let K1 be the value corresponding to the breakaway point K1 = 1/4. If we let K = 1/4 + K2 , we have the locus equation s2 + s + 1/4 + K2 = 0, or   1 2 s+ + K2 = 0. (5.38) 2 The steps for plotting this locus are, of course, the same as for any other, except that now the initial departure of the locus of Eq. (5.38) corresponds to the breakaway point of the original locus of Eq. (5.11). Applying the rule for departure angles [Eq. (5.35)] from the double pole at s = − 1/2, we find that 2φdep = −180◦ − 360◦ (l − 1), ◦

◦

φdep = −90 − 180 (l − 1), ◦

φdep = ±90 (departure angles at breakaway).

(5.39) (5.40) (5.41)

In this case, the arrival angles at s = − 1/2 are, from the original root locus, along the real axis and are clearly 0◦ and 180◦ . The complete locus for our third-order example is drawn in Fig. 5.8. It combines all the results found so far—that is, the real-axis segment, the center of the asymptotes and their angles, and the angles of departure from the poles. It is usually sufficient to draw the locus by using only Rules 1 to 3, which should be memorized. Rule 4 is sometimes useful to understand how locus segments will depart, especially if there is a pole near the jω axis. Rule 5 is sometimes useful to help interpret plots that come

5.2 Guidelines for Determining a Root Locus 6 4 Imaginary axis

Figure 5.8 Root locus for 1 L(s) = s(s2 +8s+32)

233

2 0 2 4 6 10

5

0

5

Real axis

from the computer and, as we shall see in the next section, to explain qualitative changes in some loci as a pole or zero is moved. The actual locus in Fig. 5.8 was drawn using the MATLAB commands numL = [1]; denL = [1 8 32 0]; sysL = tf(numL,denL); rlocus(sysL) We will next summarize the rules for drawing a root locus.

5.2.2 Summary of the Rules for Determining a Root Locus RULE 1. The n branches of the locus start at the poles of L(s) and m branches end on the zeros of L(s). RULE 2. The loci are on the real axis to the left of an odd number of poles and zeros. RULE 3. For large s and K, n − m of the loci are asymptotic to lines at angles φl radiating out from the center point s = α on the real axis, where 180◦ + 360◦ (l − 1) , l = 1, 2, . . . , n − m, (5.42) φl = n−m   pi − zi α= . (5.43) n−m RULE 4. The angle(s) of departure of a branch of the locus from a pole of multiplicity q is given by   qφl,dep = ψi − φi − 180◦ − 360◦ (l − 1), (5.44) and the angle(s) of arrival of a branch at a zero of multiplicity q is given by   qψl,arr = φi − ψi + 180◦ + 360◦ (l − 1).

(5.45)

RULE 5. The locus can have multiple roots at points on the locus of multiplicity q. The branches will approach a point of q roots at angles separated by 180◦ + 360◦ (l − 1) (5.46) q

234 Chapter 5 The Root-Locus Design Method and will depart at angles with the same separation, forming an array of 2q rays equally spaced. If the point is on the real axis, then the orientation of this array is given by the real-axis rule. If the point is in the complex plane, then the angle of departure rule must be applied.

5.2.3 Selecting the Parameter Value The positive root locus is a plot of all possible locations for roots to the equation 1 + KL(s) = 0 for some real positive value of K. The purpose of design is to select a particular value of K that will meet the specifications for static and dynamic response. We now turn to the issue of selecting K from a particular locus so that the roots are at specific places. Although we shall show how the gain selection can be made by hand calculations from a plot of the locus, this is almost never done by hand because the determination can be accomplished easily by MATLAB. It is useful, however, to be able to perform a rough sanity check on the computer-based results by hand. Using Definition II of the locus, we developed rules to sketch a root locus from the phase of L(s) alone. If the equation is actually to have a root at a particular place when the phase of L(s) is 180◦ , then a magnitude condition must also be satisfied. This condition is given by Eq. (5.9), rearranged as K =−

1 . L(s)

For values of s on the root locus, the phase of L(s) is 180◦ , so we can write the magnitude condition as 1 . (5.47) K= |L| Equation (5.47) has both an algebraic and a graphical interpretation. To see the latter, consider the locus of 1 + KL(s), where L(s) =

1 . s[(s + 4)2 + 16]

(5.48)

For this transfer function, the locus is plotted in Fig. 5.9. In Fig. 5.9, the lines corresponding to a damping ratio of ζ = 0.5 are sketched and the points where the locus crosses these lines are marked with dots (•). Suppose we wish to set the gain so that the roots are located at the dots. This corresponds to selecting the gain so that two of the closed-loop system poles have a damping ratio of ζ = 0.5. (We will find the third pole shortly.) What is the value of K when a root is at the dot? From Eq. (5.47), the value of K is given by 1 over the magnitude of L(s0 ), where s0 is the coordinate of the dot. On the figure we have plotted three vectors marked s0 − s1 , s0 − s2 , and s0 − s3 , which are the vectors from the poles of L(s) to the point s0 . (Since s1 = 0, the first vector equals s0 .) Algebraically, we have L(s0 ) =

1 . s0 (s0 − s2 )(s0 − s3 )

(5.49)

Using Eq. (5.47), this becomes K=

1 = |s0 ||s0 − s2 ||s0 − s3 |. |L(s0 )|

(5.50)

5.2 Guidelines for Determining a Root Locus s0 6

z

s2

0.5

4 s0 Imaginary axis

Figure 5.9 Root locus for L(s) = s[(s+4)1 2 +16] showing calculations of gain K

235

s1

2 s0

s3

z

0.5

0 2 4 6

Graphical calculation of the desired gain

10

8

6

4

2 0 Real axis

2

4

6

The graphical interpretation of Eq. (5.50) shows that its three magnitudes are the lengths of the corresponding vectors drawn on Fig. 5.9 (see Appendix WD). Hence we can compute the gain to place the roots at the dot (s = s0 ) by measuring the lengths of these vectors and multiplying the lengths together, provided that the scales of the imaginary and real axes are identical. Using the scale of the figure, we estimate that |s0 | ∼ = 4.0, ∼ |s0 − s2 | = 2.1, ∼ 7.7. |s0 − s3 | = Thus the gain is estimated to be K = 4.0(2.1)(7.7) ∼ = 65. We conclude that if K is set to the value 65, then a root of 1+KL will be at s0 , which has the desired damping ratio of 0.5. Another root is at the conjugate of s0 . Where is the third root? The third branch of the locus lies along the negative real axis. If performing the calculations by hand, we would need to take a test point, compute a trial gain, and repeat this process until we have found the point where K = 65. However, if performing a check on MATLAB’s determination, it is sufficient to merely use the procedure above to verify the gain at the root location indicated by the computer. To use MATLAB, plot the locus using the command rlocus(sysL), for example, then the command [K,p] = rlocfind(sysL) will produce a crosshair on the plot and, when spotted at the desired location of the root and selected with a mouse click, the value of the gain K is returned as well as the roots corresponding to that K in the variable p. The use of rltool makes this even easier, and will be discussed in more detail in Example 5.7. Finally, with the gain selected, it is possible to compute the error constant of the control system. A process with the transfer function given by Eq. (5.48) has one integrator and, in a unity feedback configuration, will be a Type 1 control system.

236 Chapter 5 The Root-Locus Design Method In this case the steady-state error in tracking a ramp input is given by the velocity constant: Kv = lim sKL(s) s→0

= lim s s→0

K s[(s + 4)2 + 16]

(5.51) (5.52)

K . (5.53) 32 With the gain set for complex roots at a damping ζ = 0.5, the root-locus gain is K = 65, so from Eq. (5.53) we get Kv = 65/32 ∼ = 2. If the closed-loop dynamic response, as determined by the root locations, is satisfactory and the steady-state accuracy, as measured by Kv , is good enough, then the design can be completed by gain selection alone. However, if no value of K satisfies all of the constraints, as is typically the case, then additional modifications are necessary to meet the system specifications. =

5.3 Selected Illustrative Root Loci A number of important control problems are characterized by a process with the simple “double integrator” transfer function 1 (5.54) G(s) = 2 . s The control of attitude of a satellite is described by this equation. Also, the read/write head assembly of a computer hard-disk drive is typically floating on an air bearing so that friction is negligible for all but the smallest motion. The motor is typically driven by a current source so the back emf does not affect the torque. The result is a plant described by Eq. (5.54). If we form a unity feedback system with this plant, and a proportional controller, the root locus with respect to controller gain is 1 1 + kp 2 = 0. (5.55) s If we apply the rules to this (trivial) case, the results are as follows: RULE 1. The locus has two branches that start at s = 0. RULE 2. There are no parts of the locus on the real axis. RULE 3. The two asymptotes have origin at s = 0 and are at the angles of ±90◦ . RULE 4. The loci depart from s = 0 at the angles of ±90◦ . Conclusion: The locus consists of the imaginary axis and the transient would be oscillatory for any value of kp . A more useful design results with the use of proportional plus derivative control.

EXAMPLE 5.3

Root Locus for Satellite Attitude Control with PD Control The characteristic equation with PD control is 1  1 + kp + kD s 2 = 0. s

(5.56)

237

5.3 Selected Illustrative Root Loci 3 (s+1) s2

2 Imaginary axis

Figure 5.10 Root locus for L(s) = G(s) =

1 䊊

0 1 2 3 6

4

2 Real axis

0

2

To put the equation in root-locus form, we define K = kD , and for the moment arbitrarily select the gain ratio9 as kp /kD = 1, which results in the root-locus form 1+K

s+1 s2

= 0.

(5.57)

Solution. Again we compute the results of the rules: RULE 1. There are two branches that start at s = 0, one of which terminates on the zero at s = −1 and the other of which approaches infinity. RULE 2. The real axis to the left of s = −1 is on the locus. RULE 3. Since n − m = 1, there is one asymptote along the negative real axis. RULE 4. The angles of departure from the double pole at s = 0 are ±90◦ . RULE 5. From Rules 1–4, it should be clear that the locus will curl around the zero, rejoin the real axis to the left of the zero, and terminate as indicated by Rule 1. It turns out that the locus segments rejoin the real axis at s = −2, which creates a point of multiple roots. Evaluation of the angle of arrival at this point will show that the segments arrive at ±90◦ , from which on the locus from Rule 2: it is a point of multiple roots, in this case a point of break in. We conclude that two branches of the locus leave the origin going north and south and that they curve around10 without passing into the RHP and break into the real axis at s = −2, from which point one branch goes west toward infinity and the other goes east to rendezvous with the zero at s = −1. The locus is plotted in Fig. 5.10 with the commands numS = [1 1]; denS = [1 0 0]; sysS = tf(numS,denS); rlocus( sysS)

9 Given a specific physical system, this number would be selected with consideration of the specified rise time of the design or the maximum control signal (control authority) of the actuator. 10 You can prove that the path is a circle by assuming that s + 1 = ejθ and showing that the equation has a solution for a range of positive K and real θ under this assumption. (See Problem 5.18.)

238 Chapter 5 The Root-Locus Design Method Comparing this case with that for the simple 1/s2 , we see that Effect of a Zero in the LHP

The addition of the zero has pulled the locus into the LHP, a point of general importance in constructing a compensation. In the previous case, we considered pure PD control. However, as we have mentioned earlier, the physical operation of differentiation is not practical and in practice PD control is approximated by kD s , (5.58) D(s) = kp + s/p + 1 which can be put in root-locus form by defining K = kp +pkD and z = pkp /K so that11 s+z . (5.59) D(s) = K s+p For reasons we will see when we consider design by frequency response, this controller transfer function is called a “lead compensator” or, referring to the frequent implementation by electrical components, a “lead network.” The characteristic equation for the 1/s2 plant with this controller is 1 + D(s)G(s) = 1 + KL(s) = 0, s+z = 0. 1+K 2 s (s + p)

EXAMPLE 5.4

Root Locus of the Satellite Control with Modified PD or Lead Compensation To evaluate the effect of the added pole, we will again set z = 1 and consider three different values for p. We begin with a somewhat large value, p = 12, and consider the root locus for s+1 1+K 2 . (5.60) s (s + 12) Solution. Again, we apply the rules for plotting a root locus: RULE 1. There are now three branches to the locus, two starting at s = 0 and one starting at s = −12. RULE 2. The real axis segment −12 ≤ s ≤ −1 is part of the locus. RULE 3. There are n−m = 3−1 = 2 asymptotes centered at α = and at the angles ±90◦ .

−12−(−1) 2

= − 11/2

RULE 4. The angles of departure of the branches at s = 0 are again ±90◦ . The angle of departure from the pole at s = −12 is at 0◦ . There are several possibilities on how the locus segments behave while still adhering to the guidance above. MATLAB is the expedient way to discover the paths. The MATLAB commands 11 The use of z here for zero is not to be confused with the use of the operator z used in defining the discrete transfer function needed to describe digital controllers.

5.3 Selected Illustrative Root Loci 3 2 Imaginary axis

Figure 5.11 Root locus for L(s) = s2(s+1) (s+12)

239

1 䊊

0 1 2 3 6

4

2 Real axis

0

2

numL = [1 1]; denL = [1 12 0 0]; sysL = tf(numL,denL); rlocus(sysL) show that two branches of locus break vertically from the poles at s = 0, curve around to the left without passing into the RHP, and break in at s = −2.3, where one branch goes right to meet the zero at s = −1 and the other goes left, where it is met by the root that left the pole at s = −12. These two form a multiple root at s = −5.2 and break away there and approach the vertical asymptotes located at s = −5.5. The locus is plotted in Fig. 5.11. Considering this locus, we see that the effect of the added pole has been to distort the simple circle of the PD control but, for points near the origin, the locus is quite similar to the earlier case. The situation changes when the pole is brought closer in.

EXAMPLE 5.5

Root Locus of the Satellite Control with Lead Having a Relatively Small Value for the Pole Now consider p = 4 and draw the root locus for s+1 1+K 2 = 0. s (s + 4)

(5.61)

Solution. Again, by the rules, we have the following: RULE 1. There are again three branches to the locus, two starting from s = 0 and one from s = −4. RULE 2. The segment of the real axis −4 ≤ s ≤ −1 is part of the locus. RULE 3. There are two asymptotes centered at α = − 3/2 and at the angles ±90◦ . RULE 4. The branches again depart from the poles at s = 0 at ±90◦ . RULE 5. The MATLAB commands numL=[1 1]; denL=[1 4 0 0];

240 Chapter 5 The Root-Locus Design Method 3 2 Imaginary axis

Figure 5.12 Root locus for L(s) = s2(s+1) (s+4)

1 䊊

0 1 2 3 6

4

2 Real axis

0

2

sysL=tf(numL,denL) rlocus(sysL) show that two branches of this locus break away vertically from the poles at s = 0, curve slightly to the left and join the asymptotes going north and south. The locus segment from the root at s = −4 goes east and terminates at the zero. In this case, the locus differs from the case when p = −12 in that there are no break-in or breakaway points on the real axis as part of the locus. The MATLAB plot is given in Fig. 5.12. In these two cases we have similar systems, but in one case, p = −12, there were both a break-in and a breakaway on the real axis, whereas for p = −4, these features have disappeared. A logical question might be to ask at what point they went away. As a matter of fact, it happens at p = 9, and we’ll look at that locus next.

EXAMPLE 5.6

The Root Locus for the Satellite with a Transition Value for the Pole Plot the root locus for 1+K

s+1 s (s + 9) 2

= 0.

(5.62)

Solution RULE 1. The locus has three branches, starting from s = 0 and s = −9. RULE 2. The real axis segment −9 ≤ s ≤ −1 is part of the locus. RULE 3. The two asymptotes are centered at α = − 8/2 = −4. RULE 4. The departures are, as before, at ±90◦ from s = 0. RULE 5. The MATLAB commands numL=[1 1]; denL=[1 9 0 0] sysL=tf(numL,denL); rlocus(sysL) produces the locus in Fig. 5.13. It shows the two branches of this locus break away vertically from the poles at s = 0 and curl around and join the real axis again at

5.3 Selected Illustrative Root Loci 3 2 Imaginary axis

Figure 5.13 Root locus for L(s) = s2(s+1) (s+9)

241

1 䊊

0 1 2 3 6

4

2 Real axis

0

2

s = −3 with an angle of arrival of ±60◦ while the branch from the pole at s = −9 heads east and joins the other two poles at s = −3 with an angle of arrival of 0◦ . These three locus segments continue on by splitting out of s = −3 at the departure angles of 0◦ and ±120◦ , with one heading into the zero and the other two heading away to the northwest to join the asymptotes. Using Rule 5 would confirm these angles of arrival and departure.12 From Figs. 5.11 through 5.13, it is evident that when the third pole is near the zero (p near 1), there is only a modest distortion of the locus that would result for D(s)G(s) ∼ = K s12 , which consists of two straight-line locus branches departing at ±90◦ from the two poles at s = 0. Then, as we increase p, the locus changes until at p = 9 the locus breaks in at −3 in a triple multiple root. As the pole p is moved to the left beyond −9, the locus exhibits distinct break-in and breakaway points, approaching, as p gets very large, the circular locus of one zero and two poles. Figure 5.13, when p = 9, is thus a transition locus between the two second-order extremes, which occur at p = 1 (when the zero is canceled) and p → ∞ (where the extra pole has no effect).

EXAMPLE 5.7

An Exercise to Repeat the Prior Examples Using RLTOOL Repeat Examples 5.3 through 5.6 using MATLAB’s RLTOOL feature. Solution. RLTOOL is an interactive root-locus design tool in MATLAB that provides a graphical user interface (GUI) for performing root-locus analysis and design. RLTOOL provides an easy way to design feedback controllers because it allows rapid iterations and quickly shows their effect on the resulting root locus. To illustrate the use of the tool, the MATLAB commands numL=[1 1]; denL=[1 0 0]; sysL=tf(numL,denL) rltool(sysL) 12 The shape of this special root locus is a trisectrix of Maclaurin, a plane curve that can be used to trisect an angle.

242 Chapter 5 The Root-Locus Design Method Figure 5.14 RLTOOL graphical user interface

will initiate the GUI and produce the root locus shown in Fig. 5.10, which is similar to Examples 5.4 through 5.6, but without the pole on the negative real axis that was moved around for illustration purposes in the three prior examples. By clicking on “Compensator Editor” in the “Control and Estimation Tools Manager” window, right clicking on the “Dynamics” dialog window and selecting “add pole/zero”, you can add a pole at the location s = −12. This will produce the locus that is shown in Fig. 5.11 and Fig. 5.14. Now put your mouse on the pole at s = −12, hold down the mouse button, and slide it from s = −12 to s = −4 slowly, so you can examine the locus shapes at all intermediate points. Be especially careful (and slow) as you pass through s = −9 because the locus shape changes very quickly with the pole in this region. Note that you can also put your mouse on one of the closed-loop poles (squares) and slide that along the locus. It will show you the location of the other roots that correspond to that value of the gain, K, and the frequency and damping of the closed-loop roots will be shown for when the roots are complex pairs. More detail can be found in the RLTOOL Tutorial in Appendix WR.

5.3 Selected Illustrative Root Loci

243

A useful conclusion drawn from this example is the following: An additional pole moving in from the far left tends to push the locus branches to the right as it approaches a given locus. The double integrator is the simplest model of the examples, assuming a rigid body with no friction. A more realistic case would include the effects of flexibility in the satellite attitude control, where at least the solar panels would be flexible. In the case of the disk drive read/write mechanism, the head and supporting arm assembly always has flexibility and usually a very complex behavior with a number of lightly damped modes, which can often be usefully approximated by a single dominant mode. In Section 2.1 it was shown that flexibility in the disk drive added a set of complex poles to the 1/s2 model. Generally there are two possibilities, depending on whether the sensor is on the same rigid body as the actuator, which is called the collocated case,13 or is on another body, in which case we have the noncollocated case.14 We begin with consideration of the collocated case similar to that given by Eq. (2.20). As we saw in Chapter 2, the transfer function in the collocated case has not only a pair of complex poles but also a pair of nearby complex zeros located at a lower natural frequency than the poles. The numbers in the examples that follow are chosen more to illustrate the root-locus properties than to represent particular physical models.

EXAMPLE 5.8

Root Locus of the Satellite Control with a Collocated Flexibility Plot the root locus of the characteristic equation 1 + G(s)D(s) = 0, where G(s) =

(s + 0.1)2 + 62 s [(s + 0.1)2 + 6.62 ] 2

(5.63)

is in a unity feedback structure with the controller transfer function D(s) = K

s+1 . s + 12

(5.64)

Solution. In this case L(s) =

(s + 0.1)2 + 62 s+1 s + 12 s2 [(s + 0.1)2 + 6.62 ]

has both poles and zeros near the imaginary axis and we should expect to find the departure angles of particular importance. RULE 1. There are five branches to the locus, three of which approach finite zeros and two of which approach asymptotes. RULE 2. The real-axis segment −12 ≤ s ≤ −1 is part of the locus. 13 Typical of the satellite attitude control, where the flexibility arises from solar panels and both actuator and sensor act on the main body of the satellite. 14 Typical of the satellite, where the flexibility arises from a scientific package whose attitude is to be controlled from a command body coupled to the package by a flexible strut. This case is also typical of computer hard-disk read/write head control, where the motor is on one end of the arm and the head is on the other.

244 Chapter 5 The Root-Locus Design Method

s+1 (s+0.1)2 +62 s+12 s2 (s+0.1)2 +6.62

8

f1 䊊c 1

6 4 Imaginary axis

Figure 5.15 Figure for computing a departure angle for L(s) =

2

f5

0

c3

2 4 6 8 15

f2, f3

䊊

10

5 Real axis

c2 䊊 f4 0

5

RULE 3. The center of the two asymptotes is at α=

11 −12 − 0.1 − 0.1 − (−0.1 − 0.1 − 1) =− . 5−3 2

The angle of the asymptotes is ±90◦ . RULE 4. We compute the departure angle from the pole at s = −0.1 + j6.6. The angle at this pole we will define to be φ1 . The other angles are marked on Fig. 5.15. The root-locus condition is φ1 = ψ1 + ψ2 + ψ3 − (φ2 + φ3 + φ4 + φ5 ) − 180◦ , φ1 = 90◦ + 90◦ + tan−1 (6.6) − [90◦ + 90◦ + 90◦   6.6 ] − 180◦ , + tan−1 12

(5.65)

φ1 = 81.4◦ − 90◦ − 28.8◦ − 180◦ , = −217.4◦ = 142.6◦ , so the root leaves this pole up and to the left, into the stable region of the plane. An interesting exercise would be to compute the arrival angle at the zero located at s = −0.1 + j6. Using MATLAB, the locus is plotted in Fig. 5.16. Note that all the attributes that were determined using the simple rules were exhibited by the plot, thus verifying in part that the data were entered correctly. The previous example showed that In the collocated case, the presence of a single flexible mode introduces a lightly damped root to the characteristic equation but does not cause the system to be unstable. The departure angle calculation showed that the root departs from the pole introduced by the flexible mode toward the LHP. Next, let’s consider the noncollocated

245

5.3 Selected Illustrative Root Loci

(s+0.1)2 +62 s2 (s+0.1)2 +6.62

8 䊊

6 4 Imaginary axis

Figure 5.16 Root locus for s+1 L(s) = s+12

2 䊊

0 2 4

䊊

6 8 15

10

5 Real axis

0

5

case, for which we take the plant transfer function to be G(s) =

1 s [(s + 0.1)2 + 6.62 ] 2

,

(5.66)

compensated again by the lead D(s) = K

s+1 . s + 12

(5.67)

As these equations show, the noncollocated transfer function has the complex poles but does not have the associated complex zeros as occurred in the previous example and that we also saw for the collated case of Chapter 2 in Eq. (2.20). This will have a substantial effect, as illustrated by Example 5.9.

EXAMPLE 5.9

Root Locus for the Noncollocated Case Apply the rules and draw the root locus for KL(s) = DG = K

1 s+1 , s + 12 s2 [(s + 0.1)2 + 6.62 ]

(5.68)

paying special attention to the departure angles from the complex poles. RULE 1. There are five branches to the root locus, of which one approaches the zero and four approach the asymptotes. RULE 2. The real-axis segment defined by −12 ≤ s ≤ −1 is part of the locus. RULE 3. The center of the asymptotes is located at α=

−12 − 0.2 − (−1) −11.2 = , 5−1 4

and the angles for the four asymptotic branches are at ±45◦ , ±135◦ . RULE 4. We again compute the departure angle from the pole at s = −0.1+j6.6. The angle at this pole we will define to be φ1 . The other angles are marked on Fig. 5.17.

246 Chapter 5 The Root-Locus Design Method

s+1 1 s+12 s2 (s+0.1)2 +6.62

8

f1

6 4 Imaginary axis

Figure 5.17 Figure to compute a departure angle for L(s) =

2

f5

0

f2, f3

䊊

c1

2 4

f4

6 8 15

Figure 5.18 Root locus for L(s) =

5 Real axis

0

5

0

5

8 6 4 Imaginary axis

1 s+1 s+12 s2 (s+0.1)2 +6.62

10

2 䊊

0 2 4 6 8 15

10

5 Real axis

The root locus condition is φ1 = ψ1 − (φ2 + φ3 + φ4 + φ5 ) − 180◦ ,    6.6 φ1 = tan−1 (6.6) − 90◦ + 90◦ + 90◦ + tan−1 − 180◦ , 12 φ1 = 81.4◦ − 90◦ − 90◦ − 90◦ − 28.8◦ − 180◦ , ◦

◦

◦

(5.69)

◦

φ1 = 81.4 − 90 − 28.8 − 360 , φ1 = −37.4◦ . In this case, the root leaves the pole down and to the right, toward the unstable region. We would expect the system to soon become unstable as gain is increased. RULE 5. The locus is plotted in Fig. 5.18 with the commands numG = 1; denG = [1.0 0.20 43.57 0 0]; sysG = tf(numG,denG); numD = [1 1]; denD = [1 12]; sysD = tf(numD,denD); sysL = sysD*sysG; rlocfind(sysL)

5.3 Selected Illustrative Root Loci

247

and is seen to agree with the calculations above. By using RLTOOL, we see that the locus from the complex poles enter into the RHP almost immediately as the gain is increased. Furthermore, by selecting those roots so that they are just to the left of the imaginary axis, it can be seen that the dominant slow roots down near the origin have extremely low damping. Therefore, this system will have a very lightly damped response with very oscillatory flexible modes. It would not be considered acceptable with the lead compensator as chosen for this example.

A Locus with Complex Multiple Roots We have seen loci with break-in and breakaway points on the real axis. Of course, an equation of fourth or higher order can have multiple roots that are complex. Although such a feature of a root locus is a rare event, it is an interesting curiosity that is illustrated by the next example.

EXAMPLE 5.10

Root Locus Having Complex Multiple Roots Sketch the root locus of 1 + KL(s) = 0, where L(s) =

1 . s(s + 2)[(s + 1)2 + 4]

Solution RULE 1. There are four branches of the locus, all of which approach asymptotes. RULE 2. The real-axis segment −2 ≤ s ≤ 0 is on the locus. RULE 3. The center of the asymptotes is at α=

−2 − 1 − 1 − 0 + 0 = −1 4−0

and the angles are φl = 45◦ , 135◦ , −45◦ , −135◦ . RULE 4. The departure angle φdep from the pole at = −1 + 2j, based on Fig. 5.19, is 3 f3 2 Imaginary axis

Figure 5.19 Figure to compute departure angle for 1 L(s) = s(s+2)(s+1) 2 +4

1

f1

f2

0 1

f4

2 3 4

2

0 Real axis

2

4

248 Chapter 5 The Root-Locus Design Method 3 2 Imaginary axis

Figure 5.20 Root locus for 1 L(s) = s(s+2)(s+1) 2 +4

1 0 1 2 3 4

2

0 Real axis

2

4

φdep = φ3 = −φ1 − φ2 − φ4 + 180◦     2 −1 −1 2 − tan − 90◦ + 180◦ = − tan −1 1 − 116.6◦ − 63.4◦ − 90◦ + 180◦ = −90◦ . We can observe at once that, along the line s = −1 + jω, φ2 and φ1 are angles of an isosceles triangle and always add to 180◦ . Hence, the entire line from one complex pole to the other is on the locus in this special case. RULE 5. Using MATLAB, we see that there are multiple roots at s = −1 ± 1.22j, and branches of the locus come together at −1 ± 1.22j. Using Rule 5, we can verify that the locus segments break away at 0◦ and 180◦ as shown by MATLAB. The locus in this example is a transition between two types of loci: one where the complex poles are to the left of the example case and approach the asymptotes at ±135◦ and another where the complex poles are to the right of their positions in the example and approach the asymptotes at ±45◦ .

5.4 Design Using Dynamic Compensation Consideration of control design begins with the design of the process itself. The importance of early consideration of potential control problems in the design of the process and selection of the actuator and sensor cannot be overemphasized. It is not uncommon for a first study of the control to suggest that the process itself can be changed by, for example, adding damping or stiffness to a structure to make a flexibility easier to control. Once these factors have been taken into account, the design of the controller begins. If the process dynamics are of such a nature that a satisfactory design cannot be obtained by adjustment of the proportional gain alone, then some modification or compensation of the dynamics is indicated. While the variety of possible compensation schemes is great, three categories have been found to

249

5.4 Design Using Dynamic Compensation

Lead and lag compensations

be particularly simple and effective. These are lead, lag, and notch compensations.15 Lead compensation approximates the function of PD control and acts mainly to speed up a response by lowering rise time and decreasing the transient overshoot. Lag compensation approximates the function of PI control and is usually used to improve the steady-state accuracy of the system. Notch compensation will be used to achieve stability for systems with lightly damped flexible modes, as we saw with the satellite attitude control having noncollocated actuator and sensor. In this section we will examine techniques to select the parameters of these three schemes. Lead, lag, and notch compensations have historically been implemented using analog electronics and hence were often referred to as networks. Today, however, most new control system designs use digital computer technology, in which the compensation is implemented in the software. In this case, one needs to compute discrete equivalents to the analog transfer functions, as described in Chapter 4 and discussed further in Chapter 8 and in Franklin et al. (1998). Compensation with a transfer function of the form D(s) = K

s+z s+p

(5.70)

is called lead compensation if z < p and lag compensation if z > p. Compensation is typically placed in series with the plant in the feed-forward path, as shown in Fig. 5.21. It can also be placed in the feedback path and in that location has the same effect on the overall system poles but results in different transient responses from reference inputs. The characteristic equation of the system in Fig. 5.21 is 1 + D(s)G(s) = 0, 1 + KL(s) = 0, where K and L(s) are selected to put the equation in root-locus form as before.

5.4.1 Design Using Lead Compensation To explain the basic stabilizing effect of lead compensation on a system, we first consider proportional control for which D(s) = K. If we apply this compensation to a second-order position control system with normalized transfer function G(s) =

Figure 5.21 Feedback system with compensation

R




Controller





D(s)

U

1 , s(s + 1)

Plant G(s)

Y

15 The names of these compensation schemes derive from their frequency (sinusoidal) responses, wherein the output leads the input in one case (a positive phase shift) and lags the input in another (a negative phase shift). The frequency response of the third looks as if a notch had been cut in an otherwise flat frequency response. See Chapter 6.

250 Chapter 5 The Root-Locus Design Method 3 2 Imaginary axis

Figure 5.22 Root loci for 1 + D(s)G(s) = 0, 1 : with G(s) = s(s+1) compensation D(s) = K (solid lines) and with D(s) = K(s + 2) (dashed lines)

1 䊊

0 1 2 3 6

Selection of the zero and pole of a lead

5

4

3

2 1 Real axis

0

1

2

the root locus with respect to K is shown as the solid-line portion of the locus in Fig. 5.22. Also shown in Fig. 5.22 is the locus produced by proportional plus derivative control, where D(s) = K(s+2). The modified locus is the circle sketched with dashed lines. As we saw in the examples, the effect of the zero is to move the locus to the left, toward the more stable part of the s-plane. If, now, our speed-of-response specification calls for ωn ∼ = 2, then proportional control alone (D = K) can produce only a very low value of damping ratio ζ when the roots are put at the required value of ωn . Hence, at the required gain, the transient overshoot will be substantial. However, by adding the zero of PD control we can move the locus to a position having closed-loop roots at ωn = 2 and damping ratio ζ ≥ 0.5. We have “compensated” the given dynamics by using D(s) = K(s + 2). As we observed earlier, pure derivative control is not normally practical because of the amplification of sensor noise implied by the differentiation and must be approximated. If the pole of the lead compensation is placed well outside the range of the design ωn , then we would not expect it to upset the dynamic response of the design in a serious way. For example, consider the lead compensation s+2 . D(s) = K s+p The root loci for two cases with p = 10 and p = 20 are shown in Fig. 5.23, along with the locus for PD control. The important fact about these loci is that for small gains, before the real root departing from −p approaches −2, the loci with lead compensation are almost identical to the locus for which D(s) = K(s + 2). Note that the effect of the pole is to lower the damping, but for the early part of the locus, the effect of the pole is not great if p > 10. Selecting exact values of z and p in Eq. (5.70) for particular cases is usually done by trial and error, which can be minimized with experience. In general, the zero is placed in the neighborhood of the closed-loop ωn , as determined by rise-time or settling-time requirements, and the pole is located at a distance 5 to 20 times the value of the zero location. The choice of the exact pole location is a compromise between the conflicting effects of noise suppression, for which one wants a small value for p, and compensation effectiveness for which one wants a large p. In general, if the

5.4 Design Using Dynamic Compensation Figure 5.23 Root loci for three cases 1 with G(s) = s(s+1) : (s+2) (s+20) ; (s+2) (s+10) ;

(b) D(s) = (c) D(s) = s + 2 (solid lines)

3 Lead pole at 10 2 Imaginary axis

(a) D(s) =

251

Lead pole at 20

1 䊊

0 1 PD control 2 3 6

5

4

3

2 1 Real axis

0

1

2

pole is too close to the zero, then, as seen in Fig. 5.23, the root locus moves back too far toward its uncompensated shape and the zero is not successful in doing its job. On the other hand, for reasons that are perhaps easier to understand from the frequency response, when the pole is too far to the left, the magnification of sensor noise appearing at the output of D(s) is too great and the motor or other actuator of the process can be overheated by noise energy in the control signal, u(t). With a large value of p, the lead compensation approaches pure PD control. A simple example will illustrate the approach.

EXAMPLE 5.11

Design Using Lead Compensation Find a compensation for G(s) = 1/[s(s + 1)] that will provide overshoot of no more than 20% and rise time of no more than 0.3 sec. Solution. From Chapter 3, we estimate that a damping ratio of ζ ≥ 0.5 and a natural ∼ frequency of ωn ∼ = 1.8 0.3 = 6 should satisfy the requirements. To provide some margin, we will shoot for ζ ≥ 0.5 and ωn ≥ 7 rad/sec. Considering the root loci plotted in Fig. 5.23, we will first try s+2 . D(s) = K s + 10 Figure 5.24 shows that K = 70 will yield ζ = 0.56 and ωn = 7.7 rad/sec, which satisfies the goals based the initial estimates. The third pole will be at s = −2.4 with K = 70. Because this third pole is so near the lead zero at −2, the overshoot should not be increased very much from the second-order case. However, Fig. 5.25 shows the step response of the system exceeds the overshoot specification a small amount. Typically, lead compensation in the feed-forward path will increase the stepresponse overshoot because the zero of the compensation has a differentiating effect, as discussed in Chapter 3. The rise-time specification has been met because the time for the amplitude to go from 0.1 to 0.9 is less than 0.3 sec. We want to tune the compensator to achieve better damping in order to reduce the overshoot in the transient response. The expedient way to do this is to use RLTOOL,

252 Chapter 5 The Root-Locus Design Method Figure 5.24 Root locus for lead design

8 K  70

6

Damping  0.5

Imaginary axis

4 v7 2 䊊

0 2 4 6 8 14 12 10 8

Figure 5.25 Step response for Example 5.11

6 4 2 Real axis

0

2

4

1.4 1.2

Amplitude

1 0.8 0.6 0.4 0.2 0

0

0.3

0.6

0.9 1.2 Time (sec)

1.5

1.8

sysG=tf(1,[1 1 0]); sysD=tf([1 2],[1 10]); rltool(sysG,sysD) By moving the pole of the lead compensator more to the left in order to pull the locus in that direction, and selecting K = 91, we obtain D(s) = 91

(s + 2) , (s + 13)

which will provide more damping than the previous design iteration. Figure 5.26 shows the root locus with the s-plane regions superimposed on the same plot from RLTOOL. The transient response from RLTOOL is shown in Fig. 5.27 and demonstrates that the overshoot specification is now met (in fact exceeded) with Mp = 17% and the rise time has degraded some from the previous iteration, but still satisfies the 0.3-sec specification.

5.4 Design Using Dynamic Compensation Figure 5.26 Illustration of the tuning of the dynamic lead compensator using RLTOOL

Figure 5.27 Step response for K = 91 and (s+2) 1 L(s) = (s+13) s(s+1)

1.4

1.2

Amplitude

1

0.8

0.6

0.4

0.2

0

0

0.2

0.4

0.6

0.8

1 1.2 Time (sec)

1.4

1.6

1.8

253

254 Chapter 5 The Root-Locus Design Method As stated earlier, the name lead compensation is a reflection of the fact that to sinusoidal signals, these transfer functions impart phase lead. For example, the phase of Eq. (5.70) at s = jω is given by     −1 ω −1 ω − tan . (5.71) φ = tan z p If z < p, then φ is positive, which by definition indicates phase lead. The details of design using the phase angle of the lead compensation will be treated in Chapter 6.

5.4.2 Design Using Lag Compensation Once satisfactory dynamic response has been obtained, perhaps by using one or more lead compensations, we may discover that the low-frequency gain—the value of the relevant steady-state error constant, such as Kv —is still too low. As we saw in Chapter 4, the system type, which determines the degree of the polynomial the system is capable of following, is determined by the order of the pole of the transfer function D(s)G(s) at s = 0. If the system is Type 1, the velocity-error constant, which determines the magnitude of the error to a ramp input, is given by lims→0 sD(s)G(s). In order to increase this constant, it is necessary to do so in a way that does not upset the already satisfactory dynamic response. Thus, we want an expression for D(s) that will yield a significant gain at s = 0 to raise Kv (or some other steady-state error constant) but is nearly unity (no effect) at the higher frequency ωn , where dynamic response is determined. The result is D(s) =

An example of lag compensation

s+z , s+p

z > p,

(5.72)

where the values of z and p are small compared with ωn , yet D(0) = z/p = 3 to 10 (the value depending on the extent to which the steady-state gain requires boosting). Because z > p, the phase φ given by Eq. (5.71) is negative, corresponding to phase lag. Hence a device with this transfer function is called lag compensation. The effects of lag compensation on dynamic response can be studied by looking 1 at the corresponding root locus. Again, we take G(s) = [s(s+1)] , include the lead compensation KD1 (s) = K(s+2) (s+13) that produced the locus in Fig. 5.26. With the gain of K = 91 from the previous tuned example, we find that the velocity constant is Kv = lim sKD1 G s→0

= lim s(91) s→0

=

1 s+2 s + 13 s(s + 1)

91 ∗ 2 = 14. 13

Suppose we require that Kv = 70. To obtain this, we require a lag compensation with z/p = 5 in order to increase the velocity constant by a factor of 5. This can be accomplished with a pole at p = −0.01 and a zero at z = −0.05, which keeps the values of both z and p very small so that D2 (s) would have little effect on the portions of the locus representing the dominant dynamics around ωn = 7. The result is a lag

5.4 Design Using Dynamic Compensation

−12

−10

−8

−6

−4

Im(s) 8

Im(s) 0.08

6

0.06

4

0.04

2

0.02

䊊 −2

2 Re(s)

−0.14 −0.12

−0.1

䊊 −0.08 −0.06 −0.04 −0.02

0.02

−2

−0.02

−4

−0.04

−6

−0.06

−8

−0.08

255

0.04 Re(s)

Figure 5.28 Root locus with both lead and lag compensations

compensation with the transfer function of D2 (s) = (s+0.05) (s+0.01) . The root locus with both lead and lag compensation is plotted in Fig. 5.28 and we see that, for the large scale on the left, the locus is not noticeably different from that in Fig. 5.26. This was the result of selecting very small values for the pole and zero. With K = 91, the dominant roots are at −5.8 ± j6.5. The effect of the lag compensation can be seen by expanding the region of the locus around the origin as shown on the right side of Fig. 5.28. Here we can see the circular locus that is a result of the small pole and zero. A closed-loop root remains very near the lag-compensation zero at −0.05 + 0j; therefore, the transient response corresponding to this root will be a very slowly decaying term, which will have a small magnitude because the zero will almost cancel the pole in the transfer function. Still, the decay is so slow that this term may seriously influence the settling time. Furthermore, the zero will not be present in the step response to a disturbance torque and the slow transient will be much more evident there. Because of this effect, it is important to place the lag pole–zero combination at as high a frequency as possible without causing major shifts in the dominant root locations.

5.4.3 Design Using Notch Compensation Suppose the design has been completed with lead and lag compensation given by KD(s) = 91

s + 2 s + 0.05 , s + 13 s + 0.01

(5.73)

but is found to have a substantial oscillation at about 50 rad/sec when tested, because there was an unsuspected flexibility of the noncollocated type at a natural frequency

256 Chapter 5 The Root-Locus Design Method of ωn = 50. On reexamination, the plant transfer function, including the effect of the flexibility, is estimated to be G(s) =

Gain and phase stabilization

2500 . s(s + 1)(s2 + s + 2500)

(5.74)

A mechanical engineer claims that some of the “control energy” has spilled over into the lightly damped flexible mode and caused it to be excited. In other words, as we saw from the similar system whose root locus is shown in Fig. 5.18, the very lightly damped roots at 50 rad/sec have been made even less damped or perhaps unstable by the feedback. The best method to fix this situation is to modify the structure so that there is a mechanical increase in damping. Unfortunately, this is often not possible because it is found too late in the design cycle. If it isn’t possible, how else can this oscillation be corrected? There are at least two possibilities. An additional lag compensation might lower the loop gain far enough that there is greatly reduced spillover and the oscillation is eliminated. Reducing the gain at the high frequency is called gain stabilization. If the response time resulting from gain stabilization is too long, a second alternative is to add a zero near the resonance so as to shift the departure angles from the resonant poles so as to cause the closed-loop root to move into the LHP, thus causing the associated transient to die out. This approach is called phase stabilization, and its action is similar to that of flexibility in the collocated motion control discussed earlier. Gain and phase stabilization are explained more precisely by their effect on the frequency response (Chapter 6) where these methods of stabilization will be discussed further. For phase stabilization, the result is called a notch compensation, and an example has a transfer function Dnotch (s) =

s2 + 2ζ ωo s + ωo2

. (5.75) (s + ωo )2 A necessary design decision is whether to place the notch frequency above or below that of the natural resonance of the flexibility in order to get the necessary phase. A check of the angle of departure shows that with the plant as compensated by Eq. (5.73) and the notch as given, it is necessary to place the frequency of the notch above that of the resonance to get the departure angle to point toward the LHP. Thus the compensation is added with the transfer function Dnotch (s) =

s2 + 0.8s + 3600 (s + 60)2

.

(5.76)

The gain of the notch at s = 0 has been kept at 1 so as not to change the Kv . The new root locus is shown in Fig. 5.29 and the step response is shown in Fig 5.30. Note from the step response that the oscillations are well damped, the rise-time specification is still met, but the overshoot has degraded. To rectify the increased overshoot and strictly meet all the specifications, further iteration should be carried out in order to provide more damping of the fast roots in the vicinity of ωn = 7 rad/sec. When considering notch or phase stabilization, it is important to understand that its success depends on maintaining the correct phase at the frequency of the resonance. If that frequency is subject to significant change, which is common in many cases, then the notch needs to be removed far enough from the nominal frequency in order to work for all cases. The result may be interference of the notch with the rest of the

5.4 Design Using Dynamic Compensation Figure 5.29 Root locus with lead, lag, and notch compensations

0.86

0.76

0.64

0.5

0.34

0.16

䊊

60

40

257

0.94

Imaginary axis

20 0.985 100

0

80

60

40

20

䊊䊊

−20 0.985

−40 0.94 䊊

−60 0.86 −120

Figure 5.30 Step response with lead, lag, and notch compensations

0.76 −100

−80

0.64

0.5

−60 −40 Real axis

0.34

0.16

−20

0

20

1.5

Amplitude

1

0.5

0

0

0.2

0.4

0.6

0.8 1 Time (sec)

1.2

1.4

1.6

1.8

dynamics and poor performance. As a general rule, gain stabilization is substantially more robust to plant changes than is phase stabilization. Lead compensation can be implemented using analog electronics, but digital computers are preferred.

5.4.4 Analog and Digital Implementations Lead compensation can be physically realized in many ways. In analog electronics a common method is to use an operational amplifier, an example of which is shown

258 Chapter 5 The Root-Locus Design Method Figure 5.31 Possible circuit of a lead compensation

C

Rf



Vin R1

V0

R2

in Fig. 5.31. The transfer function of the circuit in Fig. 5.31 is readily found by the methods of Chapter 2 to be Dlead (s) = −a

s+z , s+p

(5.77)

where p , if Rf = R1 + R2 , z 1 , z= R1 C R1 + R2 1 . p= · R2 R1 C

a=

If a design for D(s) is complete and a digital implementation is desired, then the technique of Chapter 4 can be used by first selecting a sampling period Ts and then making substitution of T2s z−1 z+1 for s. For example, consider the lead compensation s+2 D(s) = s+13 . Then, since the rise time is about 0.3, a sampling period of six samples 2 z−1 per rise time results in the selection of Ts = 0.05 sec. With the substitution of 0.05 z+1 for s into this transfer function, the discrete transfer function is z−1 +2 z+1 z−1 + 13 40 z+1 1.55z − 1.4 = . (5.78) 1.96z − 1 Clearing fractions and using the fact that operating on the time functions zu(kTs ) = u(kTs + Ts ), we see that Eq. (5.78) is equivalent to the formula for the controller given by U(z) = E(z)

40

1 1.55 1.4 u(kTs ) + e(kTs + Ts ) − e(kTs ). 1.96 1.96 1.96 The MATLAB commands to generate the discrete equivalent controller are u(kTs + Ts ) =

(5.79)

sysC=tf([1 2],[1 13]); sysD=c2D(sysC,0.05) Fig. 5.32 shows the SIMULINK diagram for implementing the digital controller. The result of the simulation is contained in Fig. 5.33, which shows the comparison of analog and digital control outputs, and Fig. 5.34, which shows the analog and digital control outputs.

5.4 Design Using Dynamic Compensation
 Step

s
2

91

s
13

Slider Kc

Pl control

1

1

s

s
1

Tau 1

Tau 2

259

Mux Mux 1

Mux Mux 1




1.55z−1.4

91

1.96z−1

Slider Kd

Discrete Pl control

Output

Control

1

1

s

s
1

tau 1

tau 2

Figure 5.32 SIMULINK
diagram for comparison of analog and digital control

Figure 5.33 Comparison of analog and digital control output responses

1.4 Digital controller 1.2 Continuous controller

Amplitude

1 0.8 0.6 0.4 0.2 0

0

0.5

1 Time (sec)

1.5

2

As with lead compensation, lag or notch compensation can be implemented using a digital computer and following the same procedure. However, they, too, can be implemented using analog electronics, and a circuit diagram of a lag network is given in Fig. 5.35. The transfer function of this circuit can be shown to be D(s) = −a

s+z , s+p

260 Chapter 5 The Root-Locus Design Method Figure 5.34 Comparison of analog and digital control time histories

100 80

Continuous controller

Amplitude

60 40

Digital controller

20 0

20 40 0

Figure 5.35 Possible circuit of lag compensation

0.5

1 Time (sec)

1.5

2

C

R1

R2

 Ri




where R2 , Ri R1 + R2 , z= R1 R2 C 1 . p= R1 C

a=

Usually Ri = R2 , so the high-frequency gain is unity, or a = 1, and the low-frequency 2 . increase in gain to enhance Kv or other error constant is set by k = a pz = R1R+R 2

5.5 A Design Example Using the Root Locus EXAMPLE 5.12

Control of a Small Airplane For the Piper Dakota shown in Fig. 5.36, the transfer function between the elevator input and the pitch attitude is

5.5 A Design Example Using the Root Locus

261

Figure 5.36 Autopilot design in the Piper Dakota, showing elevator and trim tab Source: Photo courtesy of Denise Freeman

G(s) =

160(s + 2.5)(s + 0.7) θ(s) = 2 , δe (s) (s + 5s + 40)(s2 + 0.03s + 0.06)

(5.80)

where θ = pitch attitude, degrees (see Fig. 10.30), δe = elevator angle, degrees. (For a more detailed discussion of longitudinal aircraft motion, refer to Section 10.3.) 1. Design an autopilot so that the response to a step elevator input has a rise time of 1 sec or less and an overshoot less than 10%. 2. When there is a constant disturbing moment acting on the aircraft so that the pilot must supply a constant force on the controls for steady flight, it is said to be out of trim. The transfer function between the disturbing moment and the attitude is the same as that due to the elevator; that is, 160(s + 2.5)(s + 0.7) θ(s) = 2 , Md (s) (s + 5s + 40)(s2 + 0.03s + 0.06)

(5.81)

262 Chapter 5 The Root-Locus Design Method Md(s) dt(s) de(s)

1










ur(s)



e(s)

KD(s)

de(s)

G(s)

u(s)











G(s)

(a)

u(s)

(b)

Figure 5.37 Block diagrams for autopilot design: (a) open loop; (b) feedback scheme excluding trim control

where Md is the moment acting on the aircraft. There is a separate aerodynamic surface for trimming, δt , that can be actuated and will change the moment on the aircraft. It is shown in the close-up of the tail in Fig. 5.36. Its influence is depicted in the block diagram shown in Fig. 5.37(a). For both manual and autopilot flight, it is desirable to adjust the trim so that there is no steady-state control effort required from the elevator (that is, so δe = 0). In manual flight, this means that no force is required by the pilot to keep the aircraft at a constant altitude, whereas in autopilot control it means reducing the amount of electrical power required and saving wear and tear on the servomotor that drives the elevator. Design an autopilot that will command the trim δt so as to drive the steady-state value of δe to zero for an arbitrary constant moment Md as well as meet the specifications in part (a). Solution 1. To satisfy the requirement that the rise time tr ≤ 1 sec, Eq. (3.60) indicates that, for the ideal second-order case, ωn must be greater than 1.8 rad/sec. And to provide an overshoot of less than 10%, Fig. (3.23) indicates that ζ should be greater than 0.6, again, for the ideal second-order case. In the design process, we can examine a root locus for a candidate for feedback compensation and then look at the resulting time response when the roots appear to satisfy the design guidelines. However, since this is a fourth-order system, the design guidelines might not be sufficient, or they might be overly restrictive. To initiate the design process, it is often instructive to look at the system characteristics with proportional feedback, that is, where D(s) = 1 in Fig. 5.37(b). The statements in MATLAB to create a root locus with respect to K and a time response for the proportional feedback case with K = 0.3 are as follows: numG = 160*conv ([1 2.5],[1 0.7]); denG = conv([1 5 40],[1 0.03 0.06]); sysG = tf(numG,denG); rlocus(sysG) K = 0.3 sysL = K*sysG sysH = tf(1,1);

5.5 A Design Example Using the Root Locus

263

[sysT] = feedback (sysL,sysH) step(sysT) The resulting root locus and time response are shown with dashed lines in Figs. 5.38 and 5.39. Notice from Fig. 5.38 that the two faster roots will always have a damping ratio ζ that is less than 0.4; therefore, proportional feedback will not be acceptable. Also, the slower roots have some effect on the time response

Im(s)

Im(s)

K  1.5 10

1.0

K  0.3 5

Proportional feedback Lead compensator

20

15

10

5

0.5

䊊 3

4

Re(s)

䊊

䊊

2

Re(s) 0.5

5 K  0.3

1.0

10 K  1.5

Figure 5.38 Root loci for autopilot design Figure 5.39 Time–response plots for autopilot design

1.5

1.0 u

Proportional feedback Lead compensator

0.5

0

0

1

2

3 Time (sec)

4

5

6

264 Chapter 5 The Root-Locus Design Method shown in Fig. 5.39 (dashed curve) with K = 0.3 in that they cause a long-term settling. However, the dominating characteristic of the response that determines whether or not the compensation meets the specifications is the behavior in the first few seconds, which is dictated by the fast roots. The low damping of the fast roots causes the time response to be oscillatory, which leads to excess overshoot and a longer settling time than desired. We saw in Section 5.4.1 that lead compensation causes the locus to shift to the left, a change needed here to increase the damping. Some trial and error will be required to arrive at a suitable pole and zero location. Values of z = 3 and p = 20 in Eq. (5.70) have a substantial effect in moving the fast branches of the locus to the left; thus D(s) =

Lead compensation via MATLAB

s+3 . s + 20

Trial and error is also required to arrive at a value of K that meets the specifications. The statements in MATLAB to add this compensation are numD = [1 3]; denD = [1 20]; sysD = tf(numD,denD); sysDG = sysD*sysG rlocus(sysDG) K = 1.5; sysKDG = K*sysDG; sysH = tf(1,1) sysT = feedback(sysKDG,sysH) step(sysT) The root locus for this case and the corresponding time response are also shown in Figs. 5.38 and 5.39 by the solid lines. Note that the damping of the fast roots that corresponds to K = 1.5 is ζ = 0.52, which is slightly lower than we would like; also, the natural frequency is ωn = 15 rad/sec, much faster than we need. However, these values are close enough to meeting the guidelines to suggest a look at the time response. In fact, the time response shows that tr ∼ = 0.9 sec and Mp ∼ = 8%, both within the specifications, although by a very slim margin. In sum, the primary design path consisted of adjusting the compensation to influence the fast roots, examining their effect on the time response, and continuing the design iteration until the time specifications were satisfied. 2. The purpose of the trim is to provide a moment that will eliminate a steady-state nonzero value of the elevator. Therefore, if we integrate the elevator command δe and feed this integral to the trim device, the trim should eventually provide the moment required to hold an arbitrary altitude, thus eliminating the need for a steady-state δe . This idea is shown in Fig. 5.40(a). If the gain on the integral term KI is small enough, the destabilizing effect of adding the integral should be small and the system should behave approximately as before, since that feedback loop has been left intact. The block diagram in Fig. 5.40(a) can be reduced to that in

5.5 A Design Example Using the Root Locus

KI s

dt

1




265

Md




 Md

ur








KD(s)

de







G(s)

u

ur






DI(s)









G(s)

u



(a)

(b)

Figure 5.40 Block diagram showing the trim-command loop Fig. 5.40(b) for analysis purposes by defining the compensation to include the PI form   KI . DI (s) = KD(s) 1 + s However, it is important to keep in mind that, physically, there will be two outputs from the compensation: δe (used by the elevator servomotor) and δt (used by the trim servomotor). The characteristic equation of the system with the integral term is KI KDG = 0. s To aid in the design process, it is desirable to find the locus of roots with respect to KI , but the characteristic equation is not in any of the root-locus forms given by Eqs. (5.6)–(5.9). Therefore, dividing by 1 + KDG yields 1 + KDG +

(KI /s)KDG = 0. 1 + KDG To put this system in root locus form, we define 1+

L(s) =

1 KDG . s 1 + KDG

(5.82)

KDG In MATLAB, with 1+KGD already computed as sysT, we construct the integrator as sysIn = tf(1,[1 0]), the loop gain of the system with respect to KI as sysL = sysIn*sysT, and the root locus with respect to KI is found with rltool(sysL). It can be seen from the locus in Fig. 5.41 that the damping of the fast roots decreases as KI increases, as is typically the case when integral control is added. This shows the necessity for keeping the value of KI as low as possible. After some trial and error, we select KI = 0.15. This value has little effect on the roots—note the roots are virtually on top of the previous roots obtained without the integral term—and little effect on the short-term behavior of the step response, as shown in Fig. 5.42(a), so the specifications are still met. KI = 0.15 does cause the longer-term attitude behavior to approach the commanded value with no error, as we would expect with integral control. It also causes δe to

266 Chapter 5 The Root-Locus Design Method Im(s) 15 Im(s) 10

1.5 1.0

5 0.5

20

15

10

5

Re(s)

4

䊊 3

䊊

2

1

䊊

Re(s) 0.5

5 1.0 10

1.5

15

Figure 5.41 Root locus versus KI : assumes an added integral term and lead compensation with a gain K = 1.5; roots for KI = 0.15 marked with • approach zero [Fig. 5.42(b) shows it settling in approximately 30 sec], which is good because this is the reason for choosing integral control in the first place. The time for the integral to reach the correct value is predicted by the new, slow real root that is added by the integral term at s = −0.14. The time constant associated with this root is τ = 1/0.14 ∼ = 7 sec. The settling time to 1% for a root with σ = 0.14 is shown by Eq. (3.65) to be ts = 33 sec, which agrees with the behavior in Fig. 5.42(b).

5.6 Extensions of the Root-Locus Method As we have seen in this chapter, the root-locus technique is a graphical scheme to show locations of possible roots of an algebraic equation as a single real parameter varies. The method can be extended to consider negative values of the parameter, a sequential consideration of more than one parameter, and systems with time delay. In this section we examine these possibilities. Another interesting extension to nonlinear systems is in Chapter 9.

5.6.1 Rules for Plotting a Negative (0◦ ) Root Locus We now consider modifying the root-locus procedure to permit analysis of negative values of the parameter. In a number of important cases, the transfer function of the

5.6 Extensions of the Root-Locus Method Figure 5.42 Step response for the case with an integral term and 5◦ command

267

6 5

(°)

4 3 2 1 0

0

5

10

15 Time (sec)

20

25

30

0

5

10

15 Time (sec)

20

25

30

0.2

e (°)

0.1

0

0.1 0.2

plant has a zero in the RHP and is said to be nonminimum phase. The result is often a locus of the form 1 + A(zi − s)G
(s) = 1 + (−A)(s − zi )G
(s) = 0, and in the standard form the parameter K = −A must be negative. Another important issue calling for understanding the negative locus arises in building a control system. In any physical implementation of a control system there are inevitably a number of amplifiers and components whose gain sign must be selected. By Murphy’s Law,16 when the loop is first closed, the sign will be wrong and the behavior will be unexpected unless the engineer understands how the response will go if the gain which should be positive is instead negative. So what are the rules for a negative locus (a root locus relative to a negative parameter)? First of all, Eqs. (5.6)–(5.9) must be satisfied for negative values of K, which implies that L(s) is real and positive. In other words, for the negative locus, the phase condition is Definition of a Negative Root Locus

The angle of L(s) is 0◦ + 360◦ (l − 1) for s on the negative locus. The steps for plotting a negative locus are essentially the same as for the positive locus, except that we search for places where the angle of L(s) is 0◦ + 360◦ (l − 1) 16Anything that can go wrong, will go wrong.

268 Chapter 5 The Root-Locus Design Method instead of 180◦ + 360◦ (l − 1). For this reason, a negative locus is also referred to as a 0◦ root locus. This time we find that the locus is to the left of an even number of real poles plus zeros (the number zero being even). Computation of the asymptotes for large values of s is, as before, given by   pi − z i α= , (5.83) n−m but we modify the angles to be φl = (shifted by locus:

180◦ (n−m)

360◦ (l − 1) , l = 1, 2, 3, . . . , n − m n−m

from the 180◦ locus). Following are the guidelines for plotting a 0◦

RULE 1. (As before) The n branches of the locus leave the poles and m approach the zeros and n − m approach asymptotes to infinity. RULE 2. The locus is on the real axis to the left of an even number of real poles plus zeros. RULE 3. The asymptotes are described by   −a1 + b1 p i − zi = , α= n−m n−m 360◦ (l − 1) φl = , l = 1, 2, 3, . . . , n − m. n−m Notice that the angle condition here is measured from 0◦ rather than from 180◦ as it was in the positive locus. RULE 4. Departure angles from poles and arrival angles to zeros are found by searching in the near neighborhood of the pole or zero where the phase of L(s) is 0◦ , so that   ψi − φi − 360◦ (l − 1), qφdep =   qψarr = φi − ψi + 360◦ (l − 1), where q is the order of the pole or zero and l takes on q integer values such that the angles are between ±180◦ . RULE 5. The locus can have multiple roots at points on the locus and the branches will approach a point of q roots at angle separated by 180◦ + 360◦ (l − 1) q and will depart at angles with the same separation. The result of extending the guidelines for constructing root loci to include negative parameters is that we can visualize the root locus as a set of continuous curves showing the location of possible solutions to the equation 1 + KL(s) = 0 for all real values of K, both positive and negative. One branch of the locus departs from every pole in one direction for positive values of K, and another branch departs from

5.6 Extensions of the Root-Locus Method

269

the same pole in another direction for negative K. Likewise, all zeros will have two branches arriving, one with positive and the other with negative values of K. For the n − m excess poles, there will be 2(n − m) branches of the locus asymptotically approaching infinity as K approaches positive and negative infinity, respectively. For a single pole or zero, the angles of departure or arrival for the two locus branches will be 180◦ apart. For a double pole or zero, the two positive branches will be 180◦ apart and the two negative branches will be at 90◦ to the positive branches. The negative locus is often required when studying a nonminimum phase transfer function. A well-known example is that of the control of liquid level in the boiler of a steam power plant. If the level is too low, the actuator valve adds (relatively) cold water to the boiling water in the vessel. The initial effect of the addition is to slow down the rate of boiling, which reduces the number and size of the bubbles and causes the level to fall momentarily before the added volume and heat cause it to rise again to the new increased level. This initial underflow is typical of nonminimum phase systems. Another typical nonminimum phase transfer function is that of the altitude control of an airplane. To make the plane climb, the upward deflection of the elevators initially causes the plane to drop before it rotates and climbs. A Boeing 747 in this mode can be described by the scaled and normalized transfer function 6−s G(s) = . (5.84) 2 s(s + 4s + 13) To put 1 + KG(s) in root-locus form, we need to multiply by −1 to get s−6 G(s) = − 2 . (5.85) s(s + 4s + 13)

EXAMPLE 5.13

Negative Root Locus for an Airplane Sketch the negative root locus for the equation s−6 1+K 2 = 0. s(s + 4s + 13)

(5.86)

Solution RULE 1. There are three branches and two asymptotes. RULE 2. A real-axis segment is to the right of s = 6 and a segment is to the left of s = 0. RULE 3. The angles of the asymptotes are φl = of the asymptotes is at α = −2−2−(6) = −5. 3−1

(l−1)360◦ 2

= 0◦ , 180◦ , and the center

RULE 4. The branch departs the pole at s = −2 + j3 at the angle     3 3 −1 −1 φ = tan − tan − 90◦ + 360◦ (l − 1), −6 −2 φ = 159.4 − 123.7 − 90 + 360◦ (l − 1), φ = −54.3◦ . The locus is plotted in Fig. 5.43 by MATLAB, which is seen to be consistent with these values.

270 Chapter 5 The Root-Locus Design Method 6 4 Imaginary axis

Figure 5.43 Negative root locus corresponding to L(s) = (s − 6)/s(s2 + 4s + 13)

2 䊊

0 2 4 6 5

0

5

10

Real axis


5.6.2 Consideration of Two Parameters Successive loop closure

EXAMPLE 5.14

An important technique for practical control is to consider a structure with two loops, an inner loop around an actuator or part of the process dynamics and an outer loop around the entire plant-plus-inner-controller. The process is called successive loop closure. A controller is selected for the inner loop to be robust and give good response alone, and then the outer loop can be designed to be simpler and more effective than if the entire control was done without the aid of the inner loop. The use of the root locus to study such a system with two parameters can be illustrated by a simple example.

Root Locus Using Two Parameters in Succession A block diagram of a relatively common servomechanism structure is shown in Fig. 5.44. Here a speed-measuring device (a tachometer) is available and the problem is to use the root locus to guide the selection of the tachometer gain KT as well as the amplifier gain KA . The characteristic equation of the system in Fig. 5.44 is 1+

KA KT + = 0, s(s + 1) s + 1

which is not in the standard 1+KL(s) form. After clearing fractions, the characteristic equation becomes s2 + s + KA + KT s = 0, (5.87) which is a function of two parameters, whereas the root locus technique can consider only one parameter at a time. In this case, we set the gain KA to a nominal value of 4 Figure 5.44 Block diagram of a servomechanism structure, including tachometer feedback

R








KA








1 s
1 KT

1

1 s

Y

5.6 Extensions of the Root-Locus Method

271

and consider first the locus with respect to KT . With KA = 4, Eq. (5.87) can be put s , or into root-locus form for a root-locus study with respect to KT with L(s) = s2 +s+4 1 + KT

s2

s = 0. +s+4

(5.88)

For this root locus, the zero is at s = 0 and the poles are at the roots of s2 + s + 4 = 0, or s = − 21 ±1.94j. A sketch of the locus using the rules as before is shown in Fig. 5.45. From this locus, we can select KT so the complex roots have a specific damping ratio or take any other value of KT that would result in satisfactory roots for the characteristic equation. Consider KT = 1. Having selected a trial value of KT , we can now re-form the equation to consider the effects of changing from KA = 4 by taking the new parameter to be K1 so that KA = 4 + K1 . The locus with respect to 1 , so that the locus is for the K1 is governed by Eq. (5.50), now with L(s) = s2 +2s+4 equation 1 1 + K1 2 = 0. (5.89) s + 2s + 4 Note that the poles of the new locus corresponding to Eq. (5.89) are the roots of the previous locus, which was drawn versus KT , and the roots were taken at KT = 1. The locus is sketched in Fig. 5.46, with the previous locus versus KT left dashed. We Figure 5.45 Root locus of closed-loop poles of the system in Fig. 5.44 versus KT

Im(s)

䊊

Figure 5.46 Root locus versus K1 = KA + 4 after choosing KT = 1

Re(s)

Im(s)

䊊

Re(s)

272 Chapter 5 The Root-Locus Design Method could draw a locus with respect to K1 for a while, stop, resolve the equation, and continue the locus with respect to KT , in a sort of see-saw between the parameters KA and KT , and thus use the root locus to study the effects of two parameters on the roots of a characteristic equation. Notice, of course, that we can also plot the root locus for negative values of K1 and thus consider values of KA less than 4.


5.6.3 Time Delay

Time delays always reduce the stability of a system.

An example of a root locus with time delay

Time delays often arise in control systems, both from delays in the process itself and from delays in the processing of sensed signals. Chemical plants often have processes with a time delay representing the time material takes to be transported via pipes or other conveyer. In measuring the attitude of a spacecraft en route to Mars, there is a significant time delay for the sensed quantity to arrive back on Earth due to the speed of light. There is also a small time delay in any digital control system due to the cycle time of the computer and the fact that data is processed at discrete intervals. Time delay always reduces the stability of a system; therefore, it is important to be able to analyze its effect. In this section we discuss how to use the root locus for such analysis. Although an exact method of analyzing time delay is available in the frequency-response methods to be described in Chapter 6, knowing several different ways to analyze a design provides the control designer with more flexibility and an ability to check the candidate solutions. Consider the problem of designing a control system for the temperature of the heat exchanger described in Chapter 2. The transfer function between the control As and the measured output temperature Tm is described by two first-order terms plus a time delay Td of 5 sec. The time delay results because the temperature sensor is physically located downstream from the exchanger, so that there is a delay in its reading. The transfer function is G(s) =

e−5s , (10s + 1)(60s + 1)

(5.90)

where the e−5s term arises from the time delay.17 The corresponding root-locus equations with respect to proportional gain K are 1 + KG(s) = 0, 1+K

e−5s (10s + 1)(60s + 1)

= 0,

600s2 + 70s + 1 + Ke−5s = 0.

(5.91)

How would we plot the root locus corresponding to Eq. (5.91)? Since it is not a polynomial, we cannot proceed with the methods used in previous examples. So we reduce the given problem to one we have previously solved by approximating the nonrational function e−5s with a rational function. Since we are concerned with control systems and hence typically with low frequencies, we want an approximation that will be good 17 Time delay is often referred to as “transportation lag” in the process industries.

5.6 Extensions of the Root-Locus Method

Padé approximant

Contrasting methods of approximating delay

Figure 5.47 Poles and zeros of the Padé approximants to e−s , with superscripts identifying the corresponding approximants; for example, x 1 represents the (1, 1) approximant

273

for small s.18 The most common means for finding such an approximation is attributed to H. Padé. It consists of matching the series expansion of the transcendental function e−5s with the series expansion of a rational function whose numerator is a polynomial of degree p and whose denominator is a polynomial of degree q. The result is called a (p, q) Padé approximant19 to e−5s . We will initially compute the approximants to e−s , and in the final result we will substitute Td s for s to allow for any desired delay. The resulting (1, 1) Padé approximant (p = q = 1) is (see Appendix W5 for details) 1 − (Td s/2) . (5.92) e−Td s ∼ = 1 + (Td s/2) If we assume p = q = 2, we have five parameters and a better match is possible. In this case we have the (2, 2) approximant, which has the transfer function 1 − Td s/2 + (Td s)2 /12 e−Td s ∼ . (5.93) = 1 + Td s/2 + (Td s)2 /12 The comparison of these approximants can be seen from their pole–zero configurations as plotted in Fig. 5.47. The locations of the poles are in the LHP and the zeros are in the RHP at the reflections of the poles. In some cases a very crude approximation is acceptable. For small delays the (0, 1) approximant can be used, which is simply a first-order lag given by 1 ∼ e−Td s = . (5.94) 1 + Td s To illustrate the effect of a delay and the accuracy of the different approximations, root loci for the heat exchanger are drawn in Fig. 5.48 for four cases. Notice that, for low gains and up to the point where the loci cross the imaginary axis, the approximate curves are very close to the exact. However, the (2, 2) Padé curve follows the exact curve much further than does the first-order lag, and its increased accuracy would be useful if the delay were larger. All analyses of the delay show its destabilizing effect and how it limits the achievable response time of the system. Im(s) 2

2 䊊

2 1 1

3

2

1

1

1 䊊 2

3

Re(s)

1 2

2

2 䊊

18 The nonrational function e−5s is analytic for all finite values of s and so may be approximated by a

√ rational function. If nonanalytic functions such as s were involved, great caution would be needed in selecting an approximation valid near s = 0. 19 The (p, p) Padé approximant for a delay of T seconds is most commonly used and is computed by the MATLAB command [num,den] = pade(T, P).

274 Chapter 5 The Root-Locus Design Method Im(s) 0.6 0.4

䊊

0.2 No delay Delay approximated by first-order lag, Eq. (5.94) Delay approximated by Padé (2, 2) approximant, Eq. (5.93) Delay  e5s (exact)

0.5

0

0.5

Re(s)

0.2 0.4

䊊

0.6

Figure 5.48 Root loci for the heat exchanger with and without time delay While the Padé approximation leads to a rational transfer function, in theory it is not necessary for plotting a root locus. A direct application of the phase condition can be used to plot portions of an exact locus of a system with time delay. The phase-angle condition does not change if the transfer function of the process is nonrational, so we still must search for values of s for which the phase is 180◦ + 360◦ l. If we write the transfer function as ¯ G(s) = e−Td s G(s), ¯ the phase of G(s) is the phase of G(s) minus λω for s = σ + jω. Thus we can ¯ formulate a root-locus problem as searching for locations where the phase of G(s) is ◦ ◦ 180 + Td ω + 360 (l − 1). To plot such a locus, we would fix ω and search along a horizontal line in the s-plane until we found a point on the locus, then raise the value of ω, change the target angle, and repeat. Similarly, the departure angles are modified by Td ω, where ω is the imaginary part of the pole from which the departure is being computed. MATLAB does not provide a program to plot the root locus of systems with delay, so we must be satisfied here with Padé approximants. Since it is possible to plot the frequency response (or Bode plot) of delay exactly and easily, if the designer feels that the Padé approximant is not satisfactory, the expedient approach is to use the frequency-response design methods described in Chapter 6.

5.7 Historical Perspective In Chapter 1 we gave an overview of the early development of feedback control analysis and design including frequency response and root-locus design. Root-locus design was introduced in 1948 by Walter R. Evans, who was working in the field of guidance and control of aircraft and missiles at the Autonetics Division of North American Aviation (now a part of The Boeing Co.). Many of his problems involved unstable or neutrally stable dynamics, which made the frequency methods difficult, so he suggested returning to the study of the characteristic equation that had been the

5.7 Historical Perspective

275

basis of the work of Maxwell and Routh nearly 70 years earlier. However, rather than treat the algebraic problem, Evans posed it as a graphical problem in the complex s-plane. Evans was also interested in the character of the dynamic response of the aerospace vehicles being controlled; therefore he wanted to solve for the closed loop roots in order to understand the dynamic behavior. To facilitate this understanding, Evans developed techniques and rules allowing one to follow graphically the paths of the roots of the characteristic equation as a parameter was changed. His method is suitable for design as well as for stability analysis and remains an important technique today. Originally, it enabled the solutions to be carried out by hand since computers were not available to design engineers during the 1940s; however, they remain an important tool today for aiding the design process. As we learned in this chapter, Evans method involves finding a locus of points where the angles to the other poles and zeros add up to a certain value. To aid in this determination, Evans invented the “Spirule” that is shown in Fig. 5.49. The device could be used to measure the angles and to perform the addition or subtraction very quickly. A skilled controls engineer could evaluate whether the angle criterion was met for a fairly complex design problem in a few seconds. In addition, the spiral curve on the rectangular portion of the device allowed the designer to multiply distances in order to determine the gain at a selected spot on the locus in a manner analogous to a slide rule. Evans was clearly motivated to aid the engineer who had no access to a computer in their design and analysis of control systems. Computers were virtually nonexistent in the 1940s. Large mainframe computers started being used somewhat for large-scale data processing by corporations in the 1950s, but there were no courses in engineering programs that taught the use of computers for analysis and design until about 1960. Engineering usage became commonplace through the 1960s, but the process involved submitting a job to a mainframe computer via a large deck of punched cards and waiting for the results for hours or overnight, a situation that was not conducive to any kind of design iteration. Mainframe computers in that era were just transitioning from vacuum tubes to transistors, random access memory would be in the neighborhood of 32k!, and the long-term data storage was by a magnetic tape drive. Random access drums and disks arrived during that decade, thus greatly speeding up the process of retrieving data. A big step forward in computing for engineers occurred when the batch processing based on punched cards was replaced by time share with many users at remote terminals during the late 1960s and early 1970s. Mechanical calculators were also available through the 1940s, 1950s and 1960s that could add, subtract, multiply, and divide and cost about $2000 in 1960. The very high-end devices could Figure 5.49 A Spirule: used to sketch a root locus before computers Source: Photo courtesy of David Powell

276 Chapter 5 The Root-Locus Design Method Figure 5.50 The Frieden mechanical calculator Source: Courtesy of the Computer History Museum

also do square roots. These machines were the basis for the complex computations done at Los Alamos during World War II. They were the size of a typewriter, had a large carriage that went back and forth during the calculations, and would occasionally ring a bell at the end of the carriage stroke (see Fig. 5.50). They were accurate to eight or more decimal places and were often used after the advent of computers to perform spot checks of the results, but a square root could take tens of seconds to complete, the machines were noisy, and the process was tedious. Enterprising engineers learned which particular calculations played certain tunes and it was not unusual to hear favorites, such as Jingle Bells. The personal computer arrived in the late 1970s, although the ones at that time utilized an audio cassette tape for data storage and had very limited random access memory, usually less than 16k. But as these desktop machines matured over the ensuing decade, the age of the computer for engineering design came into its own. First came the floppy disk for long-term data storage, followed by the hard drive toward the mid- and late-1980s. Initially, the BASIC and APL languages was the primary methods of programming. MATLAB was introduced by Cleve Moler in the 1970s. Two things happened in 1984: Apple introduced the point and click MacIntosh and PC-MATLAB was introduced by The Mathworks, which was specifically founded to commercialize MATLAB on personal computers. Initially, The Mathworks’ MATLAB was primarily written for control system analysis, but has branched out into many fields since the initial introduction. At that point in the evolution, the engineer could truly perform design iterations with little or no time between trials. Other similar programs were available for mainframe computers before that time; two being CTRL-C and MATRIXx; however, those programs have not adapted to the personal computer revolution and are fading from general use.

SUMMARY •

A root locus is a graph of the values of s that are solutions to the equation 1 + KL(s) = 0 with respect to a real parameter K. 1. When K > 0, s is on the locus if ∠L(s) = 180◦ , producing a 180◦ or positive K locus.

Summary

277

2. When K < 0, s is on the locus if ∠L(s) = 0◦ , producing a 0◦ or negative K locus. •

If KL(s) is the loop transfer function of a system with negative feedback, then the characteristic equation of the closed-loop system is 1 + KL(s) = 0,

• • •

and the root-locus method displays the effect of changing the gain K on the closed-loop system roots. A specific locus for a system sysL in MATLAB notation can be plotted by rlocus(sysL) and rltool(sysL) A working knowledge of how to determine a root locus is useful for verifying computer results and for suggesting design alternatives. The key features for the aid in sketching a 180◦ locus are as follows: 1. The locus is on the real axis to the left of an odd number of poles plus zeros. 2. Of the n branches, m approach the zeros of L(s) and n−m branches approach asymptotes centered at α and leaving at angles φl : n = number of poles, m = number of zeros, n − m = number of asymptotes,   pi − zi , α= n−m 180◦ + 360◦ (l − 1) , l = 1, 2, . . . , n − m. φl = n−m 3. Branches of the locus depart from the poles of order q and arrive at the zeros of order q with angles ⎞ ⎛  1 ⎝ φi − 180◦ − 360◦ (l − 1)⎠ , φl,dep = ψi − q i =l ⎛ ⎞  1 ⎝ ◦ ◦ φi − ψl,arr = ψi + 180 + 360 (l − 1)⎠ , q i =l

where q = order of the pole or zero, ψi = angles from the zeros, φi = angles from the poles. •

The parameter K corresponding to a root at a particular point s0 on the locus can be found from 1 , K= |L(s0 )| where |L(s0 )| can be found graphically by measuring the distances from s0 to each of the poles and zeros.

278 Chapter 5 The Root-Locus Design Method • •

For a locus drawn with rlocus(sysL), the parameter and corresponding roots can be found with [K, p] = rlocfind(sysL) or with rltool. Lead compensation, given by s+z , z < p, D(s) = s+p

•

approximates proportional–derivative (PD) control. For a fixed error coefficient, it generally moves the locus to the left and improves the system damping. Lag compensation, given by s+z D(s) = , z > p, s+p



•



•

approximates proportional–integral (PI) control. It generally improves the steady-state error for fixed speed of response by increasing the low-frequency gain and typically degrades stability. The root locus can be used to analyze successive loop closures by studying two (or more) parameters in succession. The root locus can be used to approximate the effect of time delay.

REVIEW QUESTIONS 1.

Give two definitions for the root locus.

2.

Define the negative root locus.

3. Where are the sections of the (positive) root locus on the real axis? 4. What are the angles of departure from two coincident poles at s = −a on the real axis? There are no poles or zeros to the right of −a. 5. What are the angles of departure from three coincident poles at s = −a on the real axis? There are no poles or zeros to the right of −a. 6. What is the principal effect of a lead compensation on a root locus? 7. What is the principal effect of a lag compensation on a root locus in the vicinity of the dominant closed-loop roots? 8. What is the principal effect of a lag compensation on the steady-state error to a polynomial reference input? 9. Why is the angle of departure from a pole near the imaginary axis especially important? 10. Define a conditionally stable system. 11. Show, with a root-locus argument, that a system having three poles at the origin MUST be conditionally stable.

PROBLEMS Problems for Section 5.1: Root Locus of a Basic Feedback System 5.1

Set up the listed characteristic equations in the form suited to Evans’s root-locus method. Give L(s), a(s), and b(s) and the parameter K in terms of the original parameters in each case. Be sure to select K so that a(s) and b(s) are monic in each case and the degree of b(s) is not greater than that of a(s).

Problems (a) (b) (c)

279

s + (1/τ ) = 0 versus parameter τ s2 + cs + c + 1 = 0 versus parameter c (s + c)3 + A(Ts + 1) = 0 (i) versus parameter A, (ii) versus parameter T , (iii) versus the parameter c, if possible. Say why you can or cannot. Can a plot of the roots be drawn versus c for given constant values of A and T by any means at all?

c(s) Ds (d) 1 + kp + ksI + τks+1 G(s) = 0. Assume that G(s) = A d(s) , where c(s) and d(s) are monic polynomials with the degree of d(s) greater than that of c(s).

(i) versus kp (ii) versus kI (iii) versus kD (iv) versus τ

Problems for Section 5.2: Guidelines for Sketching a Root Locus 5.2

Roughly sketch the root loci for the pole–zero maps as shown in Fig. 5.51 without the aid of a computer. Show your estimates of the center and angles of the asymptotes, a rough evaluation of arrival and departure angles for complex poles and zeros, and the loci for positive values of the parameter K. Each pole–zero map is from a characteristic equation of the form b(s) 1+K = 0, a(s) where the roots of the numerator b(s) are shown as small circles ◦ and the roots of the denominator a(s) are shown as ×’s on the s-plane. Note that in Fig. 5.51(c) there are two poles at the origin.

Figure 5.51 Pole–zero maps 䊊

䊊

(a)

2

䊊

(b)

䊊 䊊

(c)

䊊

(d)

(e)

(f)

280 Chapter 5 The Root-Locus Design Method 5.3

For the characteristic equation 1+ (a) (b) (c) (d)

5.4

K = 0, s(s + 1)(s + 5)

Draw the real-axis segments of the corresponding root locus. Sketch the asymptotes of the locus for K → ∞. Sketch the locus? Verify your sketch with a MATLAB plot.

Real poles and zeros. Sketch the root locus with respect to K for the equation 1+ KL(s) = 0 and the listed choices for L(s). Be sure to give the asymptotes, and the arrival and departure angles at any complex zero or pole. After completing each hand sketch, verify your results using MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. (s + 2) (a) L(s) = s(s + 1)(s + 5)(s + 10) (b) L(s) =

1 s(s + 1)(s + 5)(s + 10)

(c) L(s) =

(s + 2)(s + 6) s(s + 1)(s + 5)(s + 10)

(s + 2)(s + 4) s(s + 1)(s + 5)(s + 10) Complex poles and zeros. Sketch the root locus with respect to K for the equation 1 + KL(s) = 0 and the listed choices for L(s). Be sure to give the asymptotes and the arrival and departure angles at any complex zero or pole. After completing each hand sketch, verify your results using MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. 1 (a) L(s) = 2 s + 3s + 10 1 (b) L(s) = s(s2 + 3s + 10)

(d) L(s) = 5.5

2 (c) L(s) = (s2 + 2s + 8) s(s + 2s + 10) 2 (d) L(s) = (s 2 + 2s + 12) s(s + 2s + 10) 2 (e) L(s) = (s 2 + 1) s(s + 4)

5.6

2 (f) L(s) = (s 2 + 4) s(s + 1) Multiple poles at the origin. Sketch the root locus with respect to K for the equation 1 + KL(s) = 0 and the listed choices for L(s). Be sure to give the asymptotes and the arrival and departure angles at any complex zero or pole. After completing each hand sketch, verify your results using MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. (a) L(s) = 2 1 s (s + 8)

(b) L(s) = 3 1 s (s + 8)

Problems

281

(c) L(s) = 4 1 s (s + 8) (d) L(s) = 2(s + 3) s (s + 8) (e) L(s) = 3(s + 3) s (s + 4) + 1)2 (f) L(s) = (s s3 (s + 4)

5.7

2 (g) L(s) = 3(s + 1) 2 s (s + 10) Mixed real and complex poles. Sketch the root locus with respect to K for the equation 1 + KL(s) = 0 and the listed choices for L(s). Be sure to give the asymptotes and the arrival and departure angles at any complex zero or pole. After completing each hand sketch, verify your results using MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. (s + 2) (a) L(s) = s(s + 10)(s2 + 2s + 2)

(s + 2) (b) L(s) = 2 s (s + 10)(s2 + 6s + 25) (s + 2)2 (c) L(s) = 2 s (s + 10)(s2 + 6s + 25) 2 (d) L(s) = 2(s + 2)(s 2+ 4s + 68) s (s + 10)(s + 4s + 85) 2 (e) L(s) = 2[(s + 1) + 1] s (s + 2)(s + 3)

5.8

RHP and zeros. Sketch the root locus with respect to K for the equation 1 + KL(s) = 0 and the listed choices for L(s). Be sure to give the asymptotes and the arrival and departure angles at any complex zero or pole. After completing each hand sketch, verify your results using MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. +2 1 (a) L(s) = ss+ 10 s2 − 1 ; the model for a case of magnetic levitation with lead compensation. 1 (b) L(s) = s + 2 ; the magnetic levitation system with integral control s(s + 10) (s2 − 1) and lead compensation. (c) L(s) = s −2 1 s s2 + 2s + 1 . What is the largest value that can be obtained for (d) L(s) = s(s + 20)2 (s2 − 2s + 2) the damping ratio of the stable complex roots on this locus? (s + 2) (e) L(s) = s(s − 1)(s + 6)2 1 (f) L(s) = (s − 1)[(s + 2)2 + 3]

5.9

Put the characteristic equation of the system shown in Fig. 5.52 in root-locus form with respect to the parameter α, and identify the corresponding L(s), a(s), and b(s). Sketch

282 Chapter 5 The Root-Locus Design Method Figure 5.52 Control system for Problem 5.9

R








5 s(s
2)

Y

1
as

the root locus with respect to the parameter α, estimate the closed-loop pole locations, and sketch the corresponding step responses when α = 0, 0.5, and 2. Use MATLAB to check the accuracy of your approximate step responses. 5.10 Use the MATLAB function rltool to study the behavior of the root locus of 1 + KL(s) for L(s) =

(s + a) s(s + 1)(s2 + 8s + 52)

as the parameter a is varied from 0 to 10, paying particular attention to the region between 2.5 and 3.5. Verify that a multiple root occurs at a complex value of s for some value of a in this range. 5.11 Use Routh’s criterion to find the range of the gain K for which the systems in Fig. 5.53 are stable, and use the root locus to confirm your calculations.

R






K



s2
s
2 s(s
5)(s
6)(s2
2s
1)




Y R



s
2 s(s  2)(s2
2s
10)

K



Y

(b)

(a)

Figure 5.53 Feedback systems for Problem 5.11 5.12 Sketch the root locus for the characteristic equation of the system for which L(s) =

(s + 2) , s(s + 1)(s + 5)

and determine the value of the root-locus gain for which the complex conjugate poles have a damping ratio of 0.5. 5.13 For the system in Fig. 5.54, (a) (b) (c) (d)

Figure 5.54 Feedback system for Problem 5.13

R








Find the locus of closed-loop roots with respect to K. Is there a value of K that will cause all roots to have a damping ratio greater than 0.5? Find the values of K that yield closed-loop poles with the damping ratio ζ = 0.707. Use MATLAB to plot the response of the resulting design to a reference step.

K

(

s
1 s
13

)

s2
81
100)

s2(s2

Y

Problems

283

5.14 For the feedback system shown in Fig. 5.55, find the value of the gain K that results in dominant closed-loop poles with a damping ratio ζ = 0.5.

Figure 5.55 Feedback system for Problem 5.14




R(s)



10










1 s

1 s






Y(s)

Ks




Problems for Section 5.3: Selected Illustrative Root Loci 5.15 A simplified model of the longitudinal motion of a certain helicopter near hover has the transfer function 9.8(s2 − 0.5s + 6.3) G(s) = (s + 0.66)(s2 − 0.24s + 0.15) and the characteristic equation 1 + D(s)G(s) = 0. Let D(s) = kp at first. (a) Compute the departure and arrival angles at the complex poles and zeros. (b) Sketch the root locus for this system for parameter K = 9.8kp . Use axes −4 ≤ x ≤ 4; −3 ≤ y ≤ 3. (c) Verify your answer using MATLAB. Use the command axis([−4 4 −3 3]) to get the right scales. (d) Suggest a practical (at least as many poles as zeros) alternative compensation D(s) that will at least result in a stable system. 5.16

Figure 5.56 Control system for Problem 5.16

R

(a) For the system given in Fig. 5.56, plot the root locus of the characteristic equation as the parameter K1 is varied from 0 to ∞ with λ = 2. Give the corresponding L(s), a(s), and b(s). (b) Repeat part (a) with λ = 5. Is there anything special about this value? (c) Repeat part (a) for fixed K1 = 2, with the parameter K = λ varying from 0 to ∞.






5

K1 s
l

2 s
10 0.1

Y

0.2






1 s












5.17 For the system shown in Fig. 5.57, determine the characteristic equation and sketch the root locus of it with respect to positive values of the parameter c. Give L(s), a(s), and b(s), and be sure to show with arrows the direction in which c increases on the locus.

284 Chapter 5 The Root-Locus Design Method Figure 5.57 Control system for Problem 5.17

R








c
16s c
s

9 s2

Y

5.18 Suppose you are given a system with the transfer function L(s) =

(s + z) , (s + p)2

where z and p are real and z > p. Show that the root locus for 1 + KL(s) = 0 with respect to K is a circle centered at z with radius given by r = (z − p). Hint: Assume s + z = rejφ and show that L(s) is real and negative for real φ under this assumption. 5.19 The loop transmission of a system has two poles at s = −1 and a zero at s = −2. There is a third real-axis pole p located somewhere to the left of the zero. Several different root loci are possible, depending on the exact location of the third pole. The extreme cases occur when the pole is located at infinity or when it is located at s = −2. Give values for p and sketch the three distinct types of loci. 5.20 For the feedback configuration of Fig. 5.58, use asymptotes, center of asymptotes, angles of departure and arrival, and the Routh array to sketch root loci for the characteristic equations of the listed feedback control systems versus the parameter K. Use MATLAB to verify your results. 1 s+2 (a) G(s) = , H(s) = s(s + 1 + 3j)(s + 1 − 3j) s+8 1 s+1 H(s) = (b) G(s) = 2 , s+3 s s+7 (s + 5) , H(s) = (c) G(s) = (s + 1) s+3 (s + 3 + 4j)(s + 3 − 4j) , H(s) = 1 + 3s (d) G(s) = s(s + 1 + 2j)(s + 1 − 2j)

Figure 5.58 Feedback system for Problem 5.20

R






G(s)

Y

 H(s)

5.21 Consider the system in Fig. 5.59. (a) Using Routh’s stability criterion, determine all values of K for which the system is stable. (b) Use MATLAB to draw the root locus versus K and find the values of K at the imaginary-axis crossings.

Problems Figure 5.59 Feedback system for Problem 5.21

R






K



s
3 s(s2
4s
5)

285

Y

1 s
1

Problems for Section 5.4: Design Using Dynamic Compensation 5.22 Let

s+a 1 and D(s) = K . (s + 2)(s + 3) s+b Using root-locus techniques, find values for the parameters a, b, and K of the compensation D(s) that will produce closed-loop poles at s = −1 ± j for the system shown in Fig. 5.60. G(s) =

Figure 5.60 Unity feedback system for Problems 5.22 to 5.28 and 5.33

R






D(s)

G(s)

Y



5.23 Suppose that in Fig. 5.60 G(s) =

1 s(s2 + 2s + 2)

and

D(s) =

K . s+2

Sketch the root locus with respect to K of the characteristic equation for the closed-loop system, paying particular attention to points that generate multiple roots if KL(s) = D(s)G(s). 5.24 Suppose the unity feedback system of Fig. 5.60 has an open-loop plant given by G(s) = 1/s2 . Design a lead compensation D(s) = K s+z to be added in series with the plant so s+p that the dominant poles of the closed-loop system are located at s = −2 ± 2j. 5.25 Assume that the unity feedback system of Fig. 5.60 has the open-loop plant G(s) =

1 . s(s + 3)(s + 6)

Design a lag compensation to meet the following specifications: • • •

The step response settling time is to be less than 5 sec. The step response overshoot is to be less than 17%. The steady-state error to a unit-ramp input must not exceed 10%.

5.26 A numerically controlled machine tool positioning servomechanism has a normalized and scaled transfer function given by G(s) =

1 . s(s + 1)

Performance specifications of the system in the unity feedback configuration of Fig. 5.60 √ are satisfied if the closed-loop poles are located at s = −1 ± j 3. (a) Show that this specification cannot be achieved by choosing proportional control alone, D(s) = kp .

286 Chapter 5 The Root-Locus Design Method s+z that will meet the specification. (b) Design a lead compensator D(s) = K s+p

5.27 A servomechanism position control has the plant transfer function G(s) =

10 . s(s + 1)(s + 10)

You are to design a series compensation transfer function D(s) in the unity feedback configuration to meet the following closed-loop specifications: • • •

The response to a reference step input is to have no more than 16% overshoot. The response to a reference step input is to have a rise time of no more than 0.4 sec. The steady-state error to a unit ramp at the reference input must be less than 0.02.

(a) Design a lead compensation that will cause the system to meet the dynamic response specifications. (b) If D(s) is proportional control, D(s) = kp , what is the velocity constant Kv ? (c) Design a lag compensation to be used in series with the lead you have designed to cause the system to meet the steady-state error specification. (d) Give the MATLAB plot of the root locus of your final design. (e) Give the MATLAB response of your final design to a reference step. 5.28 Assume that the closed-loop system of Fig. 5.60 has a feed-forward transfer function G(s) =

1 . s(s + 2)

Design a lag compensation so that the dominant poles of the closed-loop system are located at s = −1 ± j and the steady-state error to a unit-ramp input is less than 0.2. 5.29 An elementary magnetic suspension scheme is depicted in Fig. 5.61. For small motions near the reference position, the voltage e on the photo detector is related to the ball displacement x (in meters) by e = 100x. The upward force (in newtons) on the ball caused by the current i (in amperes) may be approximated by f = 0.5i + 20x. The mass of the ball is 20 g and the gravitational force is 9.8 N/kg. The power amplifier is a voltage-to-current device with an output (in amperes) of i = u + V0 . (a) (b) (c) (d)

Figure 5.61 Elementary magnetic suspension

Write the equations of motion for this set-up. Give the value of the bias V0 that results in the ball being in equilibrium at x = 0. What is the transfer function from u to e? Suppose that the control input u is given by u = −Ke. Sketch the root locus of the closed-loop system as a function of K.

V0

i u

Solenoid

x e

Ball Photo detector

Light

Problems

287

s+z (e) Assume that a lead compensation is available in the form U E = D(s) = K s+p . Give values of K, z, and p that yield improved performance over the one proposed in part (d).

5.30 A certain plant with the nonminimum phase transfer function 4 − 2s G(s) = 2 s +s+9 is in a unity positive feedback system with the controller transfer function D(s). (a) Use MATLAB to determine a (negative) value for D(s) = K so that the closed-loop system with negative feedback has a damping ratio ζ = 0.707. (b) Use MATLAB to plot the system’s response to a reference step. 5.31 Consider the rocket-positioning system shown in Fig. 5.62. (a) Show that if the sensor that measures x has a unity transfer function, the lead compensator s+2 H(s) = K s+4 stabilizes the system. (b) Assume that the sensor transfer function is modeled by a single pole with a 0.1 sec time constant and unity DC gain. Using the root-locus procedure, find a value for the gain K that will provide the maximum damping ratio.

Figure 5.62 Block diagram for rocket-positioning control system

Fc






F

1 s2

x Sensor

H(s)

5.32 For the system in Fig. 5.63, (a) Find the locus of closed-loop roots with respect to K. (b) Find the maximum value of K for which the system is stable. Assume K = 2 for the remaining parts of this problem. (c) What is the steady-state error (e = r − y) for a step change in r? (d) What is the steady-state error in y for a constant disturbance w1 ?

Figure 5.63 Control system for Problem 5.32

W2

R






K




W1








1 s2 1









s









100 (s
6
2j)(s
6  2j)

Y

288 Chapter 5 The Root-Locus Design Method (e) What is the steady-state error in y for a constant disturbance w2 ? (f) If you wished to have more damping, what changes would you make to the system? 5.33 Consider the plant transfer function bs + k G(s) = 2 s [mMs2 + (M + m)bs + (M + m)k] to be put in the unity feedback loop of Fig. 5.60. This is the transfer function relating the input force u(t) and the position y(t) of mass M in the noncollocated sensor and actuator problem. In this problem we will use root-locus techniques to design a controller D(s) so that the closed-loop step response has a rise time of less than 0.1 sec and an overshoot of less than 10%. You may use MATLAB for any of the following questions: (a) Approximate G(s) by assuming that m ∼ = 0, and let M = 1, k = 1, b = 0.1, and D(s) = K. Can K be chosen to satisfy the performance specifications? Why or why not? (b) Repeat part (a) assuming that D(s) = K(s + z), and show that K and z can be chosen to meet the specifications. (c) Repeat part (b), but with a practical controller given by the transfer function D(s) = K

p(s + z) . s+p

Pick p so that the values for K and z computed in part (b) remain more or less valid. (d) Now suppose that the small mass m is not negligible, but is given by m = M/10. Check to see if the controller you designed in part (c) still meets the given specifications. If not, adjust the controller parameters so that the specifications are met. 5.34 Consider the Type 1 system drawn in Fig. 5.64. We would like to design the compensation D(s) to meet the following requirements: (1) The steady-state value of y due to a constant unit disturbance w should be less than 45 , and (2) the damping ratio ζ = 0.7. Using root-locus techniques, (a) Show that proportional control alone is not adequate. (b) Show that proportional–derivative control will work. (c) Find values of the gains kp and kD for D(s) = kp + kD s that meet the design specifications.

Figure 5.64 Control system for Problem 5.34

W

R








D(s)









1 s(s
1)

Y

Problems

289

Problems for Section 5.5: A Design Example Using the Root Locus 5.35 Consider the positioning servomechanism system shown in Fig. 5.65, where ei = Kpot θi ,

eo = Kpot θo ,

Kpot = 10 V/rad,

T = motor torque = Kt ia , km = Kt = torque constant = 0.1 N·m/A, Ra = armature resistance = 10 , Gear ratio = 1 : 1, JL + Jm = total inertia = 10−3 kg·m2 , C = 200 µF, va = KA (ei − ef ). (a) What is the range of the amplifier gain KA for which the system is stable? Estimate the upper limit graphically using a root-locus plot. (b) Choose a gain KA that gives roots at ζ = 0.7. Where are all three closed-loop root locations for this value of KA ?

Figure 5.65 Positioning servomechanism

Input potentiometer

JL

uo

ui

Source: Reprinted from Clark, 1962, with permission

Kpot

Output potentiometer

ei

ya






Jm km

KA Amplifier

Gears

Kpot

Motor eo 1000 

ef C

1000 

Feedback filter

5.36 We wish to design a velocity control for a tape-drive servomechanism. The transfer function from current I(s) to tape velocity (s) (in millimeters per millisecond per ampere) is (s) 15(s2 + 0.9s + 0.8) . = I(s) (s + 1)(s2 + 1.1s + 1) We wish to design a Type 1 feedback system so that the response to a reference step satisfies tr ≤ 4 msec, ts ≤ 15 msec, Mp ≤ 0.05.

290 Chapter 5 The Root-Locus Design Method (a) Use the integral compensator kI /s to achieve Type 1 behavior, and sketch the root locus with respect to kI . Show on the same plot the region of acceptable pole locations corresponding to the specifications. (b) Assume a proportional-integral compensator of the form kp (s + α)/s, and select the best possible values of kp and α you can find. Sketch the root-locus plot of your design, giving values for kp and α, and the velocity constant Kv your design achieves. On your plot, indicate the closed-loop poles with a dot (•) and include the boundary of the region of acceptable root locations. 5.37 The normalized, scaled equations of a cart as drawn in Fig. 5.66 of mass mc holding an inverted uniform pendulum of mass mp and length  with no friction are θ¨ − θ = −v,

(5.95)

y¨ + βθ = v, 3m

p where β = 4(m +m is a mass ratio bounded by 0 < β < 0.75. Time is measured c p)

3g(m +m ) in terms of τ = ωo t where ωo2 = (4mc +mp ) . The cart motion y is measured in units c p of pendulum length as y = 3x 4 and the input is force normalized by the system weight v = g(m u+m ) . These equations can be used to compute the transfer functions c

p

1  , =− 2 V s −1

(5.96)

s2 − 1 + β Y . = 2 2 V s (s − 1)

(5.97)

In this problem you are to design a control for the system by first closing a loop around the pendulum, Eq. (5.96), and then, with this loop closed, closing a second loop around the cart plus pendulum, Eq. (5.97). For this problem, let the mass ratio be mc = 5mp . (a) Draw a block diagram for the system with V input and both Y and  as outputs. s+z for the  loop to cancel the pole at (b) Design a lead compensation D(s) = Kp s+p s = −1 and place the two remaining poles at −4 ± j4. The new control is U(s), where the force is V (s) = U(s) + D(s)(s). Draw the root locus of the angle loop. (c) Compute the transfer function of the new plant from U to Y with D(s) in place. (d) Design a controller Dc (s) for the cart position with the pendulum loop closed. Draw the root locus with respect to the gain of Dc (s). (e) Use MATLAB to plot the control, cart position, and pendulum position for a unit step change in cart position.

Figure 5.66 Figure of cart pendulum for Problem 5.37

u

y

Trolley or cart

Problems

291

5.38 Consider the 270-ft U.S. Coast Guard cutter Tampa (902) shown in Fig. 5.67. Parameter identification based on sea-trials data (Trankle, 1987) was used to estimate the hydrodynamic coefficients in the equations of motion. The result is that the response of the heading angle of the ship ψ to rudder angle δ and wind changes w can be described by the second-order transfer functions Gδ (s) =

−0.0184(s + 0.0068) ψ(s) = , δ(s) s(s + 0.2647)(s + 0.0063)

Gw (s) =

0.0000064 ψ(s) = , w(s) s(s + 0.2647)(s + 0.0063)

where ψ = heading angle, rad, r = reference heading angle, rad, r = yaw rate, rad/sec, δ = rudder angle, rad, w = wind speed, m/sec. (a) Determine the open-loop settling time of r for a step change in δ. (b) In order to regulate the heading angle ψ, design a compensator that uses ψ and the measurement provided by a yaw-rate gyroscope (that is, by ψ˙ = r). The settling time of ψ to a step change in ψr is specified to be less than 50 sec, and for a 5◦ change in heading, the maximum allowable rudder angle deflection is specified to be less than 10◦ . (c) Check the response of the closed-loop system you designed in part (b) to a wind gust disturbance of 10 m/sec. (Model the disturbance as a step input.) If the steady-state value of the heading due to this wind gust is more than 0.5◦ , modify your design so that it meets this specification as well.

COAST GUARD

0 10 20 30 40 50

Figure 5.67 USCG cutter Tampa (902) for Problem 5.38

292 Chapter 5 The Root-Locus Design Method 5.39 Golden Nugget Airlines has opened a free bar in the tail of their airplanes in an attempt to lure customers. In order to automatically adjust for the sudden weight shift due to passengers rushing to the bar when it first opens, the airline is mechanizing a pitchattitude autopilot. Figure 5.68 shows the block diagram of the proposed arrangement. We will model the passenger moment as a step disturbance Mp (s) = M0/s, with a maximum expected value for M0 of 0.6. (a) What value of K is required to keep the steady-state error in θ to less than 0.02 rad (∼ = 1◦ )? (Assume the system is stable.) (b) Draw a root locus with respect to K. (c) Based on your root locus, what is the value of K when the system becomes unstable? (d) Suppose the value of K required for acceptable steady-state behavior is 600. Show that this value yields an unstable system with roots at s = −2.9, −13.5, +1.2 ± 6.6j. (e) You are given a black box with rate gyro written on the side and told that, when ˙ with output KT θ˙ . Assume that K = 600 installed, it provides a perfect measure of θ, as in part (d) and draw a block diagram indicating how you would incorporate the rate gyro into the autopilot. (Include transfer functions in boxes.) (f) For the rate gyro in part (e), sketch a root locus with respect to KT . (g) What is the maximum damping factor of the complex roots obtainable with the configuration in part (e)? (h) What is the value of KT for part (g)? (i) Suppose you are not satisfied with the steady-state errors and damping ratio of the system with a rate gyro in parts (e) through (h). Discuss the advantages and disadvantages of adding an integral term and extra lead networks in the control law. Support your comments using MATLAB or with rough root-locus sketches.

Figure 5.68 Golden Nugget Airlines autopilot

Mp

Elevator servo ur








K s
10

Me







Aircraft dynamics s
3 s2
4s
5

u

1 s

u

Rate gyro Attitude sensor

KTu

KT

1

5.40 Consider the instrument servomechanism with the parameters given in Fig. 5.69. For each of the following cases, draw a root locus with respect to the parameter K, and indicate the location of the roots corresponding to your final design: (a) Lead network: Let H(s) = 1,

D(s) = K

s+z , s+p

p = 6. z

Problems Figure 5.69 Control system for Problem 5.40

293

Compensator R








D(s)

u

1 s

1 s2
51s
550

Y

Sensor H(s)

Select z and K so that the roots nearest the origin (the dominant roots) yield ζ ≥ 0.4,

−σ ≤ −7,

2 Kv ≥ 16 sec. 3

(b) Output-velocity (tachometer) feedback: Let H(s) = 1 + KT s

and

D(s) = K.

Select KT and K so that the dominant roots are in the same location as those of part (a). Compute Kv . If you can, give a physical reason explaining the reduction in Kv when output derivative feedback is used. (c) Lag network: Let s+1 . H(s) = 1 and D(s) = K s+p Using proportional control, it is possible to obtain a Kv = 12 at ζ = 0.4. Select K and p so that the dominant roots correspond to the proportional-control case but with Kv = 100 rather than Kv = 12.

Problems for Section 5.6: Extensions of the Root Locus Method 5.41 Plot the loci for the 0◦ locus or negative K for each of the following: (a) (b) (c) (d) (e) (f)

The examples given in Problem 5.3 The examples given in Problem 5.4 The examples given in Problem 5.5 The examples given in Problem 5.6 The examples given in Problem 5.7 The examples given in Problem 5.8

5.42 Suppose you are given the plant 1 L(s) = 2 , s + (1 + α)s + (1 + α) where α is a system parameter that is subject to variations. Use both positive and negative root-locus methods to determine what variations in α can be tolerated before instability occurs. 5.43 Consider the system in Fig. 5.70. (a) Use Routh’s criterion to determine the regions in the (K1 , K2 ) plane for which the system is stable. (b) Use RLTOOL to verify your answer to part (a).

294 Chapter 5 The Root-Locus Design Method Figure 5.70 Feedback system for Problem 5.43

R






E





(

K1 1


K2 s

)

U

1 s(s
1)(s
0.5)

Y

5.44 The block diagram of a positioning servomechanism is shown in Fig. 5.71. (a) Sketch the root locus with respect to K when no tachometer feedback is present (KT = 0). (b) Indicate the root locations corresponding to K = 16 on the locus of part (a). For these locations, estimate the transient-response parameters tr , Mp , and ts . Compare your estimates to measurements obtained using the step command in MATLAB. (c) For K = 16, draw the root locus with respect to KT . (d) For K = 16 and with KT set so that Mp = 0.05 (ζ = 0.707), estimate tr and ts . Compare your estimates to the actual values of tr and ts obtained using MATLAB. (e) For the values of K and KT in part (d), what is the velocity constant Kv of this system?

Figure 5.71 Control system for Problem 5.44

R






K










1 s
2

1 s

Y

KT



5.45 Consider the mechanical system shown in Fig. 5.72, where g and a0 are gains. The feedback path containing gs controls the amount of rate feedback. For a fixed value of a0 , adjusting g corresponds to varying the location of a zero in the s-plane. (a) With g = 0 and τ = 1, find a value for a0 such that the poles are complex. (b) Fix a0 at this value, and construct a root locus that demonstrates the effect of varying g.

Figure 5.72 Control system for Problem 5.45

ui




1 s(ts
1)

a0





uo





gs



5.46 Sketch the root locus with respect to K for the system in Fig. 5.73 using the Pad´e(1,1) approximation and the first-order lag approximation. For both approximations, what is the range of values of K for which the system is unstable?

Problems Figure 5.73 Control system for Problem 5.46

R








K

es

1 s
0.01

295

Y



5.47 Prove that the plant G(s) = 1/s3 cannot be made unconditionally stable if pole cancellation is forbidden.



5.48 For the equation 1 + KG(s), where G(s) =

1 , s(s − p)[(s + 1)2 + 4]

use MATLAB to examine the root locus as a function of K for p in the range from p = 1 to p = 10, making sure to include the point p = 2.

6

The Frequency-Response Design Method

A Perspective on the Frequency-Response Design Method The design of feedback control systems in industry is probably accomplished using frequency-response methods more often than any other. Frequencyresponse design is popular primarily because it provides good designs in the face of uncertainty in the plant model. For example, for systems with poorly known or changing high-frequency resonances, we can temper the feedback compensation to alleviate the effects of those uncertainties. Currently, this tempering is carried out more easily using frequency-response design than with any other method. Another advantage of using frequency response is the ease with which experimental information can be used for design purposes. Raw measurements of the output amplitude and phase of a plant undergoing a sinusoidal input excitation are sufficient to design a suitable feedback control. No intermediate processing of the data (such as finding poles and zeros or determining system matrices) is required to arrive at the system model. The wide availability of computers has rendered this advantage less important now than it was years ago; however, for relatively simple systems, frequency response is often Photo courtesy of Cirrus Design Corporation

296

6.1 Frequency Response

297

still the most cost-effective design method. The method is most effective for systems that are stable in open loop. Yet another advantage is that it is the easiest method to use for designing compensation. A simple rule can be used to provide reasonable designs with a minimum of trial and error. Although the underlying theory is somewhat challenging and requires a rather broad knowledge of complex variables, the methodology of frequencyresponse design is easy, and the insights gained by learning the theory are well worth the struggle.

Chapter Overview The chapter opens with a discussion of how to obtain the frequency response of a system by analyzing its poles and zeros. An important extension of this discussion is how to use Bode plots to graphically display the frequency response. In Sections 6.2 and 6.3 we discuss stability briefly and then in more depth the use of the Nyquist stability criterion. In Sections 6.4 through 6.6 we introduce the notion of stability margins, discuss Bode’s gain–phase relationship, and study the closed-loop frequency response of dynamic systems. The gain– phase relationship suggests a very simple rule for compensation design: Shape the frequency-response magnitude so that it crosses magnitude 1 with a slope of −1. As with our treatment of the root-locus method, we describe how adding dynamic compensation can adjust the frequency response (Section 6.7) and improve system stability and/or error characteristics. We also show how to implement compensation digitally in an example. In optional Sections 6.7.7 and 6.7.8 we discuss issues of sensitivity that relate to the frequency response, including material on sensitivity functions and stability robustness. The next two sections on analyzing time delays in the system and Nichols charts represents additional, somewhat advanced material that may also be considered optional. The final Section 6.10 is a short history of the Frequency Response design method.

6.1 Frequency Response

Frequency response

The basic concepts of frequency response were discussed in Section 3.1.2. In this section we will review those ideas and extend the concepts for use in control system design. A linear system’s response to sinusoidal inputs—called the system’s frequency response—can be obtained from knowledge of its pole and zero locations. To review the ideas, we consider a system described by Y (s) = G(s), U(s) where the input u(t) is a sine wave with an amplitude A: u(t) = A sin(ωo t)1(t). This sine wave has a Laplace transform Aωo . U(s) = 2 s + ωo2

298 Chapter 6 The Frequency-Response Design Method

Partial-fraction expansion

With zero initial conditions, the Laplace transform of the output is Aωo . (6.1) Y (s) = G(s) 2 s + ωo2 A partial-fraction expansion of Eq. (6.1) [assuming that the poles of G(s) are distinct] will result in an equation of the form α1 αo αo∗ α2 αn Y (s) = + + ··· + + + , (6.2) s − p1 s − p2 s − pn s + jωo s − jωo where p1 , p2 , . . . , pn are the poles of G(s), αo would be found by performing the partial-fraction expansion, and αo∗ is the complex conjugate of αo . The time response that corresponds to Y (s) is y(t) = α1 ep1 t + α2 ep2 t + · · · + αn epn t + 2|αo | cos(ωo t + φ), t ≥ 0, (6.3) where   Im(αo ) φ = tan−1 . Re(αo ) If all the poles of the system represent stable behavior (the real parts of p1 , p2 , . . . , pn < 0), the natural unforced response will die out eventually, and therefore the steady-state response of the system will be due solely to the sinusoidal term in Eq. (6.3), which is caused by the sinusoidal excitation. Example 3.5 determined 1 the response of the system G(s) = (s+1) to the input u = sin 10t and showed that response in Fig. 3.4, which is repeated here as Fig. 6.1. It shows that e−t , the natural part of the response associated with G(s), disappears after several time constants, and the pure sinusoidal response is essentially all that remains. Example 3.5 showed that the remaining sinusoidal term in Eq. (6.3) can be expressed as y(t) = AM cos(ωo t + φ), (6.4) where  M = |G( jωo )| = |G(s)|s=jωo = {Re[G( jωo )]}2 + {Im[G( jωo )]}2 ,   Im[G( jωo )] φ = tan−1 = ∠G( jωo ). Re[G( jωo )]

Figure 6.1 Response of 1 G(s) = (s+1) to sin 10t

0.20 0.15

Output, y

0.10 0.05 0 0.05 0.10

0

1

2

3

4

5 6 Time (sec)

7

8

9

10

(6.5) (6.6)

6.1 Frequency Response

299

In polar form, G( jωo ) = Mejφ .

Frequency-response plot

Magnitude and phase

EXAMPLE 6.1

(6.7)

Equation (6.4) shows that a stable system with transfer function G(s) excited by a sinusoid with unit amplitude and frequency ωo will, after the response has reached steady state, exhibit a sinusoidal output with a magnitude M(ωo ) and a phase φ(ωo ) at the frequency ωo . The facts that the output y is a sinusoid with the same frequency as the input u and that the magnitude ratio M and phase φ of the output are independent of the amplitude A of the input are a consequence of G(s) being a linear constant system. If the system being excited were a nonlinear or time-varying system, the output might contain frequencies other than the input frequency, and the output–input ratio might be dependent on the input magnitude. More generally, the magnitude M is given by |G( jω)|, and the phase φ is given by ∠[G( jω)]; that is, the magnitude and angle of the complex quantity G(s) are evaluated with s taking on values along the imaginary axis (s = jω). The frequency response of a system consists of these functions of frequency that tell us how a system will respond to a sinusoidal input of any frequency. We are interested in analyzing the frequency response not only because it will help us understand how a system responds to a sinusoidal input, but also because evaluating G(s) with s taking on values along the jω axis will prove to be very useful in determining the stability of a closed-loop system. As we saw in Chapter 3, the jω axis is the boundary between stability and instability; we will see in Section 6.4 that evaluating G( jω) provides information that allows us to determine closed-loop stability from the open-loop G(s).

Frequency-Response Characteristics of a Capacitor Consider the capacitor described by the equation dv , dt where v is the input and i is the output. Determine the sinusoidal steady-state response of the capacitor. i=C

Solution. The transfer function of this circuit is I(s) = G(s) = Cs, V (s) so G( jω) = Cjω. Computing the magnitude and phase, we find that M = |Cjω| = Cω

and

φ = ∠(Cjω) = 90◦ .

For a unit-amplitude sinusoidal input v, the output i will be a sinusoid with magnitude Cω, and the phase of the output will lead the input by 90◦ . Note that for this example the magnitude is proportional to the input frequency while the phase is independent of frequency.

300 Chapter 6 The Frequency-Response Design Method

EXAMPLE 6.2

Frequency-Response Characteristics of a Lead Compensator Recall from Chapter 5 [Eq. (5.70)] the transfer function of the lead compensation, which is equivalent to D(s) = K

Ts + 1 , αTs + 1

α < 1.

(6.8)

1. Analytically determine its frequency-response characteristics and discuss what you would expect from the result. 2. Use MATLAB
to plot D( jω) with K = 1, T = 1, and α = 0.1 for 0.1 ≤ ω ≤ 100, and verify the features predicted from the analysis in 1, above. Solution 1. Analytical evaluation: Substituting s = jω into Eq. (6.8), we get D( jω) = K

Tjω + 1 . αTjω + 1

From Eqs. (6.5) and (6.6) the amplitude is  M = |D| = |K| 

1 + (ωT )2

1 + (αωT )2

,

and the phase is given by φ = ∠(1 + jωT ) − ∠(1 + jαωT ) = tan−1 (ωT ) − tan−1 (αωT ). At very low frequencies the amplitude is just |K|, and at very high frequencies it is | K/α |. Therefore, the amplitude is higher at very high frequency. The phase is zero at very low frequencies and goes back to zero at very high frequencies. At intermediate frequencies, evaluation of the tan−1 (·) functions would reveal that φ becomes positive. These are the general characteristics of lead compensation. 2. Computer evaluation: A MATLAB script for frequency-response evaluation was shown for Example 3.5. A similar script for the lead compensation: num = [1 1]; den = [0.1 1]; sysD = tf(num,den); w=logspace(−1,2); [mag,phase] = bode(sysD,w);

% determines frequencies over range of interest % computes magnitude and phase over frequency range of interest

loglog(w,squeeze(mag)),grid; semilogx(w,squeeze(phase)),grid; produces the frequency-response magnitude and phase plots shown in Fig 6.2.

6.1 Frequency Response 101

20

db

Magnitude 100 101

100

101

0 102

101

102

v (rad/sec) (a)

f (°)

Figure 6.2 (a) Magnitude; (b) phase for the lead compensation in Example 6.2

301

60 50 40 30 20 10 0 101

100 v (rad/sec) (b)

The analysis indicated that the low-frequency magnitude should be K (=1) and the high-frequency magnitude should be K/α (= 10), which are both verified by the magnitude plot. The phase plot also verifies that the value approaches zero at high and low frequencies and that the intermediate values are positive.

In the cases for which we do not have a good model of the system and wish to determine the frequency-response magnitude and phase experimentally, we can excite the system with a sinusoid varying in frequency. The magnitude M(ω) is obtained by measuring the ratio of the output sinusoid to input sinusoid in the steady state at each frequency. The phase φ(ω) is the measured difference in phase between input and output signals.1 A great deal can be learned about the dynamic response of a system from knowledge of the magnitude M(ω) and the phase φ(ω) of its transfer function. In the obvious case, if the signal is a sinusoid, then M and φ completely describe the response. Furthermore, if the input is periodic, then a Fourier series can be constructed to decompose the input into a sum of sinusoids, and again M(ω) and φ(ω) can be used with each component to construct the total response. For transient inputs, our best path to understanding the meaning of M and φ is to relate the frequency response G( jω) to the transient responses calculated by the Laplace transform. For example, in Fig.3.18(b) we plotted the step response of a system having the transfer function G(s) =

1 , (s/ωn )2 + 2ζ (s/ωn ) + 1

(6.9)

1Agilent Technologies produces instruments called spectral analyzers that automate this experimental procedure and greatly speed up the process.

302 Chapter 6 The Frequency-Response Design Method

Bandwidth

for various values of ζ . These transient curves were normalized with respect to time as ωn t. In Fig. 6.3 we plot M(ω) and φ(ω) for these same values of ζ to help us see what features of the frequency response correspond to the transient-response characteristics. Specifically, Figs. 3.18(b) and 6.3 indicate the effect of damping on system time response and the corresponding effect on the frequency response. They show that the damping of the system can be determined from the transient response overshoot or from the peak in the magnitude of the frequency response [(Fig. 6.3 (a)]. Furthermore, from the frequency response, we see that ωn is approximately equal to the bandwidth—the frequency where the magnitude starts to fall off from its lowfrequency value. (We will define bandwidth more formally in the next paragraph.) Therefore, the rise time can be estimated from the bandwidth. We also see that the peak overshoot in frequency is approximately 1/2ζ for ζ < 0.5, so the peak overshoot in the step response can be estimated from the peak overshoot in the frequency response. Thus, we see that essentially the same information is contained in the frequencyresponse curve as is found in the transient-response curve. A natural specification for system performance in terms of frequency response is the bandwidth, defined to be the maximum frequency at which the output of a system will track an input sinusoid in a satisfactory manner. By convention, for the system shown in Fig. 6.4 with a sinusoidal input r, the bandwidth is the frequency of r at which the output y is attenuated to a factor of 0.707 times the input.2 Figure 6.5 depicts the idea graphically for the frequency response of the closed-loop transfer function Y (s) KG(s)  T (s) = . R(s) 1 + KG(s) The plot is typical of most closed-loop systems in that (1) the output follows the input (|T | ∼ = 1) at the lower excitation frequencies, and (2) the output ceases to follow the input (|T | < 1) at the higher excitation frequencies. The maximum value of the frequency-response magnitude is referred to as the resonant peak Mr . Bandwidth is a measure of speed of response and is therefore similar to timedomain measures such as rise time and peak time or the s-plane measure of dominantroot(s) natural frequency. In fact, if the KG(s) in Fig. 6.4 is such that the closed-loop response is given by Fig. 6.3, we can see that the bandwidth will equal the natural frequency of the closed-loop root (that is, ωBW = ωn for a closed-loop damping ratio of ζ = 0.7). For other damping ratios, the bandwidth is approximately equal to the natural frequency of the closed-loop roots, with an error typically less than a factor of 2. The definition of the bandwidth stated here is meaningful for systems that have a low-pass filter behavior, as is the case for any physical control system. In other applications the bandwidth may be defined differently. Also, if the ideal model of the system does not have a high-frequency roll-off (e.g., if it has an equal number of poles and zeros), the bandwidth is infinite; however, this does not occur in nature as nothing responds well at infinite frequencies. In many cases, the designer’s primary concern is the error in the system due to disturbances rather than the ability to track an input. For error analysis, we are 2 If the output is a voltage across a 1- resistor, the power is v2 and when |v| = 0.707, the power is reduced

by a factor of 2. By convention, this is called the half-power point.

6.1 Frequency Response 10 8 6

0.1

4

0.2 0.3 0.5

2 1 0.8 0.6

0 z  0.7

0.4

0.9

db

Magnitude

20

z  0.05

0.2 20

0.1 0.08 0.06 0.04 0.02 0.01 0.1

0.2

0.4

0.6 0.8 1 v/vn

2

4

6

40 8 10

4

6

8 10

(a) 0

30

Phase

60

z  0.05 0.1 0.2 0.3

z  0.9 0.7 0.5

90

120

150

180 0.1

0.2

0.4

0.6 0.8 1 v/vn (b)

Figure 6.3 (a) Magnitude; (b) phase of Eq. (6.9)

2

303

304 Chapter 6 The Frequency-Response Design Method




KG(s)

Y



10

20 db

Figure 6.5 Definitions of bandwidth and resonant peak

R

Amplitude ratio, T(s)

Figure 6.4 Simplified system definition

Resonant peak, Mr

0

1 0.7

3

0.1

20 Bandwidth, vBW v (rad/sec)

more interested in one of the sensitivity functions defined in Section 4.1, S(s), rather than T (s). For most open-loop systems with high gain at low frequencies, S(s) for a disturbance input will have very low values at low frequencies and grows as the frequency of the input or disturbance approaches the bandwidth. For analysis of either T (s) or S(s), it is typical to plot their response versus the frequency of the input. Either frequency response for control systems design can be evaluated using the computer, or can be quickly sketched for simple systems using the efficient methods described in the following Section 6.1.1. The methods described next are also useful to expedite the design process as well as to perform sanity checks on the computer output.

6.1.1 Bode Plot Techniques Display of frequency response is a problem that has been studied for a long time. Before computers, this was accomplished by hand; therefore, it was useful to be able to accomplish this quickly. The most useful technique for hand plotting was developed by H. W. Bode at Bell Laboratories between 1932 and 1942. This technique allows plotting that is quick and yet sufficiently accurate for control systems design. Most control systems designers now have access to computer programs that diminish the need for hand plotting; however, it is still important to develop good intuition so that you can quickly identify erroneous computer results, and for this you need the ability to perform a sanity check and in some cases to determine approximate results by hand. The idea in Bode’s method is to plot magnitude curves using a logarithmic scale and phase curves using a linear scale. This strategy allows us to plot a high-order G( jω) by simply adding the separate terms graphically, as discussed in Appendix B. This addition is possible because a complex expression with zero and pole factors can be written in polar (or phasor) form as   r1 r2 r1 ejθ1 r2 ejθ2 s1s2 = = (6.10) ej(θ1 +θ2 −θ3 −θ4 −θ5 ) . G( jω) = s3s4s5 r3 ejθ3 r4 ejθ4 r5 ejθ5 r3 r4 r5

6.1 Frequency Response

Composite plot from individual terms

(The overhead arrow indicates a phasor.) Note from Eq. (6.10) that the phases of the individual terms are added directly to obtain the phase of the composite expression, G( jω). Furthermore, because r1 r 2 |G( jω)| = , r3 r4 r5 it follows that log10 |G( jω)| = log10 r1 + log10 r2 − log10 r3 − log10 r4 − log10 r5 .

Bode plot

(6.11)

We see that addition of the logarithms of the individual terms provides the logarithm of the magnitude of the composite expression. The frequency response is typically presented as two curves; the logarithm of magnitude versus log ω and the phase versus log ω. Together these two curves constitute a Bode plot of the system. Because log10 Me jφ = log10 M + jφ log10 e,

Decibel

305

(6.12)

we see that the Bode plot shows the real and imaginary parts of the logarithm of G( jω). In communications it is standard to measure the power gain in decibels (db):3 P2 |G|db = 10 log10 . (6.13) P1 Here P1 and P2 are the input and output powers. Because power is proportional to the square of the voltage, the power gain is also given by V2 . (6.14) |G|db = 20 log10 V1 Hence we can present a Bode plot as the magnitude in decibels versus log ω and the phase in degrees versus log ω.4 In this book we give Bode plots in the form log |G| versus log ω; also, we mark an axis in decibels on the right-hand side of the magnitude plot to give you the choice of working with the representation you prefer. However, for frequency-response plots, we are not actually plotting power, and use of Eq. (6.14) can be somewhat misleading. If the magnitude data are derived in terms of log |G|, it is conventional to plot them on a log scale but identify the scale in terms of |G| only (without “log”). If the magnitude data are given in decibels, the vertical scale is linear such that each decade of |G| represents 20 db.

Advantages of Working with Frequency Response in Terms of Bode Plots Advantages of Bode plots

1. 2. 3. 4.

Dynamic compensator design can be based entirely on Bode plots. Bode plots can be determined experimentally. Bode plots of systems in series (or tandem) simply add, which is quite convenient. The use of a log scale permits a much wider range of frequencies to be displayed on a single plot than is possible with linear scales.

It is important for the control systems engineer to understand the Bode plot techniques for several reasons: This knowledge allows the engineer not only to deal 3 Researchers at Bell Laboratories first defined the unit of power gain as a bel (named for Alexander Graham

Bell, the founder of the company). However, this unit proved to be too large, and hence a decibel or db ( 1/10 of a bel) was selected as a more useful unit. The abbreviation dB is also sometimes used; however, Bode used db and we choose to follow his lead. 4 Henceforth we will drop the base of the logarithm; it is understood to be 10.

306 Chapter 6 The Frequency-Response Design Method with simple problems, but also to perform a sanity check on computer results for more complicated cases. Often approximations can be used to quickly sketch the frequency response and deduce stability, as well as to determine the form of the needed dynamic compensations. Finally, an understanding of the plotting method is useful in interpreting frequency-response data that have been generated experimentally. In Chapter 5 we wrote the open-loop transfer function in the form KG(s) = K

Bode form of the transfer function

(s − z1 )(s − z2 ) · · · (s − p1 )(s − p2 ) · · ·

(6.15)

because it was the most convenient form for determining the degree of stability from the root locus with respect to the gain K. In working with frequency response, it is more convenient to replace s with jω and to write the transfer functions in the Bode form ( jωτ1 + 1)( jωτ2 + 1) · · · (6.16) KG( jω) = Ko ( jωτa + 1)( jωτb + 1) · · · because the gain Ko in this form is directly related to the transfer-function magnitude at very low frequencies. In fact, for Type 0 systems, Ko is the gain at ω = 0 in Eq. (6.16) and is also equal to the DC gain of the system. Although a straightforward calculation will convert a transfer function in the form of Eq. (6.15) to an equivalent transfer function in the form of Eq. (6.16), note that K and Ko will not usually have the same value in the two expressions. Transfer functions can also be rewritten according to Eqs. (6.10) and (6.11). As an example, suppose that KG( jω) = Ko

jωτ1 + 1 . ( jω)2 ( jωτa + 1)

(6.17)

Then ∠KG( jω) = ∠Ko + ∠( jωτ1 + 1) − ∠( jω)2 − ∠( jωτa + 1)

(6.18)

and log |KG( jω)| = log |Ko | + log |jωτ1 + 1| − log |( jω)2 | − log |jωτa + 1|. (6.19) In decibels, Eq. (6.19) becomes |KG( jω)|db = 20 log |Ko | + 20 log |jωτ1 + 1| − 20 log |( jω)2 | − 20 log |jωτa + 1|. Classes of terms of transfer functions

(6.20)

All transfer functions for the kinds of systems we have talked about so far are composed of three classes of terms: 1. Ko ( jω)n . 2. ( jωτ + 1)±1 . ±1 
 2 jω jω + 2ζ ωn + 1 . 3. ωn First we will discuss the plotting of each individual term and how the terms affect the composite plot including all the terms; then we will discuss how to draw the composite curve.

6.1 Frequency Response

307

Slope is
1, or 20 db/decade

Figure 6.6 Magnitude of ( jω)n

60

1000

50 n 
1

40

(jv)n

30 10

n  1

db

n 
2

100

20 10 0

1

10 0.1

Class 1: singularities at the origin

Class 2: first-order term

0.1

1

10 v (rad/sec)

100

1000

20

1. Ko ( jω)n Because log Ko |( jω)n | = log Ko + n log |jω|, the magnitude plot of this term is a straight line with a slope n × (20 db per decade). Examples for different values of n are shown in Fig. 6.6. Ko ( jω)n is the only class of term that affects the slope at the lowest frequencies, because all other terms are constant in that region. The easiest way to draw the curve is to locate ω = 1 and plot log Ko at that frequency. Then draw the line with slope n through that point.5 The phase of ( jω)n is φ = n × 90◦ ; it is independent of frequency and is thus a horizontal line: −90◦ for n = −1, −180◦ for n = −2, +90◦ for n = +1, and so forth. 2. jωτ + 1 The magnitude of this term approaches one asymptote at very low frequencies and another asymptote at very high frequencies: (a) For ωτ 1, jωτ + 1 ∼ = 1. (b) For ωτ  1, jωτ + 1 ∼ = jωτ .

Break point

If we call ω = 1/τ the break point, then we see that below the break point the magnitude curve is approximately constant (= 1), while above the break point the magnitude curve behaves approximately like the class 1 term Ko ( jω). The example plotted in Fig. 6.7, G(s) = 10s+1, shows how the two asymptotes cross at the break point and how the actual magnitude curve lies above that point by a factor of 1.4 (or +3 db). (If the term were in the denominator, it would be below the break point by a factor of 0.707 or −3 db.) Note that this term will have only a small effect on the composite magnitude curve below the break point, because its value is equal to 1 (= 0 db) in this region. The slope at high frequencies is +1 5 In decibels the slopes are n × 20 db per decade or n × 6 db per octave (an octave is a change in frequency by a factor of 2).

308 Chapter 6 The Frequency-Response Design Method Figure 6.7 Magnitude plot for jωτ + 1; τ = 10

Break point

10.0

40

20

G  1.4

db

jv10
1

100

1.4 1.0

3 0 Asymptotes

0.1

Figure 6.8 Phase plot for jωτ + 1; τ = 10

0.01

0.1

1.0 10 v (rad/sec)

100

Break point

20 1000

11

Asymptote

⬔(jv10
1)

90 60 11 30 Asymptote 0 30

0.01

0.02

0.1

0.2 0.4

1

v (rad/sec)

(or +20 db per decade). The phase curve can also be easily drawn by using the following low- and high-frequency asymptotes:

Class 3: second-order term

(a) For ωτ 1, ∠1 = 0◦ . (b) For ωτ  1, ∠jωτ = 90◦ . (c) For ωτ ∼ = 1, ∠( jωτ + 1) ∼ = 45◦ . For ωτ ∼ = 1, the ∠( jω + 1) curve is tangent to an asymptote going from 0◦ at ωτ = 0.2 to 90◦ at ωτ = 5, as shown in Fig. 6.8. The figure also illustrates the three asymptotes (dashed lines) used for the phase plot and how the actual curve deviates from the asymptotes by 11◦ at their intersections. Both the composite phase and magnitude curves are unaffected by this class of term at frequencies below the break point by more than a factor of 10 because the term’s magnitude is 1 (or 0 db) and its phase is less than 5◦ . 3. [( jω/ωn )2 + 2ζ ( jω/ωn ) + 1]±1 This term behaves in a manner similar to the class 2 term, with differences in detail: The break point is now ω = ωn . The magnitude changes slope by a factor of +2 (or +40 db per decade) at the break point (and −2, or −40 db per decade, when the term is in the denominator). The phase changes by ±180◦ , and the transition through the break-point region varies with the damping ratio ζ . Figure 6.3 shows the magnitude and phase for several different damping ratios when the term is in the denominator. Note that the magnitude asymptote for frequencies above the break point has a slope of −2

6.1 Frequency Response

Peak amplitude

Composite curve

309

(or −40 db per decade), and that the transition through the break-point region has a large dependence on the damping ratio. A rough determination of this transition can be made by noting that 1 at ω = ωn |G( jω)| = (6.21) 2ζ for this class of second-order term in the denominator. If the term was in the numerator, the magnitude would be the reciprocal of the curve plotted in Fig. 6.3(a). No such handy rule as Eq. (6.21) exists for sketching in the transition for the phase curve; therefore, we would have to resort to Fig. 6.3(b) for an accurate plot of the phase. However, a very rough idea of the transition can be gained by noting that it is a step function for ζ = 0, while it obeys the rule for two firstorder (class 2) terms when ζ = 1 with simultaneous break-point frequencies. All intermediate values of ζ fall between these two extremes. The phase of a second-order term is always ±90◦ at ωn . When the system has several poles and several zeros, plotting the frequency response requires that the components be combined into a composite curve. To plot the composite magnitude curve, it is useful to note that the slope of the asymptotes is equal to the sum of the slopes of the individual curves. Therefore, the composite asymptote curve has integer slope changes at each break-point frequency: +1 for a first-order term in the numerator, −1 for a first-order term in the denominator, and ±2 for secondorder terms. Furthermore, the lowest-frequency portion of the asymptote has a slope determined by the value of n in the ( jω)n term and is located by plotting the point Ko ωn at ω = 1. Therefore, the complete procedure consists of plotting the lowestfrequency portion of the asymptote, then sequentially changing the asymptote’s slope at each break point in order of ascending frequency, and finally drawing the actual curve by using the transition rules discussed earlier for classes 2 and 3. The composite phase curve is the sum of the individual curves. Addition of the individual phase curves graphically is made possible by locating the curves so that the composite phase approaches the individual curve as closely as possible. A quick but crude sketch of the composite phase can be found by starting the phase curve below the lowest break point and setting it equal to n × 90◦ . The phase is then stepped at each break point in order of ascending frequency. The amount of the phase step is ±90◦ for a first-order term and ±180◦ for a second-order term. Break points in the numerator indicate a positive step in phase, while break points in the denominator indicate a negative phase step.6 The plotting rules so far have only considered poles and zeros in the left half-plane (LHP). Changes for singularities in the right half-plane (RHP) will be discussed at the end of the section.

Summary of Bode Plot Rules 1. Manipulate the transfer function into the Bode form given by Eq. (6.16). 2. Determine the value of n for the Ko ( jω)n term (class 1). Plot the low-frequency magnitude asymptote through the point Ko at ω = 1 with a slope of n (or n×20 db per decade). 6 This approximate method was pointed out to us by our Parisian colleagues.

310 Chapter 6 The Frequency-Response Design Method 3. Complete the composite magnitude asymptotes: Extend the low-frequency asymptote until the first frequency break point. Then step the slope by ±1 or ±2, depending on whether the break point is from a first- or second-order term in the numerator or denominator. Continue through all break points in ascending order. 4. The approximate magnitude curve is increased from the asymptote value by a factor of 1.4 (+3 db) at first-order numerator break points, and decreased by a factor of 0.707 (−3 db) at first-order denominator break points. At second-order break points, the resonant peak (or valley) occurs according to Fig. 6.3(a), using the relation |G( jω)| = 1/2ζ at denominator, (or |G( jω)| = 2ζ at numerator) break points. 5. Plot the low-frequency asymptote of the phase curve, φ = n × 90◦ . 6. As a guide, the approximate phase curve changes by ±90◦ or ±180◦ at each break point in ascending order. For first-order terms in the numerator, the change of phase is +90◦ ; for those in the denominator the change is −90◦ . For second-order terms, the change is ±180◦ . 7. Locate the asymptotes for each individual phase curve so that their phase change corresponds to the steps in the phase toward or away from the approximate curve indicated by Step 6. Each individual phase curve occurs as indicated by Fig. 6.8 or Fig. 6.3(b). 8. Graphically add each phase curve. Use grids if an accuracy of about ±5◦ is desired. If less accuracy is acceptable, the composite curve can be done by eye. Keep in mind that the curve will start at the lowest-frequency asymptote and end on the highest-frequency asymptote and will approach the intermediate asymptotes to an extent that is determined by how close the break points are to each other.

EXAMPLE 6.3

Bode Plot for Real Poles and Zeros Plot the Bode magnitude and phase for the system with the transfer function KG(s) =

2000(s + 0.5) . s(s + 10)(s + 50)

Solution 1. We convert the function to the Bode form of Eq. (6.16): KG( jω) =

2[( jω/0.5) + 1] . jω[( jω/10) + 1][( jω/50) + 1]

2. We note that the term in jω is first order and in the denominator, so n = −1. Therefore, the low-frequency asymptote is defined by the first term: KG( jω) =

2 . jω

This asymptote is valid for ω < 0.1, because the lowest break point is at ω = 0.5. The magnitude plot of this term has the slope of −1 (or −20 db per decade). We locate the magnitude by passing through the value 2 at ω = 1 even though the composite curve will not go through this point because of the break point at ω = 0.5. This is shown in Fig. 6.9(a).

6.1 Frequency Response

311

3. We obtain the remainder of the asymptotes, also shown in Fig. 6.9(a): The first break point is at ω = 0.5 and is a first-order term in the numerator, which thus calls for a change in slope of +1. We therefore draw a line with 0 slope that intersects the original −1 slope. Then we draw a −1 slope line that intersects the previous one at ω = 10. Finally, we draw a −2 slope line that intersects the previous −1 slope at ω = 50. 4. The actual curve is approximately tangent to the asymptotes when far away from the break points, a factor of 1.4 (+3 db) above the asymptote at the ω = 0.5 break point, and a factor of 0.7 (−3 db) below the asymptote at the ω = 10 and ω = 50 break points. 5. Because the phase of 2/jω is −90◦ , the phase curve in Fig. 6.9(b) starts at −90◦ at the lowest frequencies. 6. The result is shown in Fig. 6.9(c). 7. The individual phase curves, shown dashed in Fig. 6.9(b), have the correct phase change for each term and are aligned vertically so that their phase change corresponds to the steps in the phase from the approximate curve in Fig. 6.9(c). Note that the composite curve approaches each individual term. 8. The graphical addition of each dashed curve results in the solid composite curve in Fig. 6.9(b). As can be seen from the figure, the vertical placement of each individual phase curve makes the required graphical addition particularly easy because the composite curve approaches each individual phase curve in turn.

EXAMPLE 6.4

Bode Plot with Complex Poles As a second example, draw the frequency response for the system 10 . KG(s) = s[s2 + 0.4s + 4]

(6.22)

Solution. A system like this is more difficult to plot than the one in the previous example because the transition between asymptotes is dependent on the damping ratio; however, the same basic ideas illustrated in Example 6.3 apply. This system contains a second-order term in the denominator. Proceeding through the steps, we convert Eq. (6.22) to the Bode form of Eq. (6.16): 1 10 . KG(s) = 2 4 s(s /4 + 2(0.1)s/2 + 1) Starting with the low-frequency asymptote, we have n = −1 and |G( jω)| ∼ = 2.5/ω. The magnitude plot of this term has a slope of −1 (−20 db per decade) and passes through the value of 2.5 at ω = 1 as shown in Fig. 6.10(a). For the second-order pole, note that ωn = 2 and ζ = 0.1. At the break-point frequency of the poles, ω = 2, the slope shifts to −3 (−60 db per decade). At the pole break point the magnitude ratio above the asymptote is 1/2ζ = 1/0.2 = 5. The phase curve for this case starts at φ = −90◦ , corresponding to the 1/s term, falls to φ = −180◦ at ω = 2 due to the pole as shown in Fig. 6.10(b), and then approaches φ = −270◦ for higher frequencies. Because the damping is small, the stepwise approximation is a very good one. The true composite phase curve is shown in Fig. 6.10(b).

312 Chapter 6 The Frequency-Response Design Method Figure 6.9 Composite plots: (a) magnitude; (b) phase; (c) approximate phase

1000

Slope  1

40 Actual magnitude curve

Slope  0

10

db

Magnitude, G

100

60

Asymptotes

20

Slope  1 2 1

0 0.1

0.1

1.0

0.5

10

10

100 Slope  2 20

50

Break points v (rad/sec) (a) 1 s/0.5
1

0

Phase, ⬔G

1 s/10
1

90 Composite 1 s/50
1 180

0.1

1.0 10 v (rad/sec) (b)

100

1000

0 Approximate

Phase, ⬔G

30 60 90

Actual

120 150 180

0.1

0.5 1.0 10 v (rad/sec) (c)

50 100

1000

6.1 Frequency Response 1000

40 Slope  1 20

10

0

1 0.1 00.1 0.1

db

60

100 Magnitude, G

Figure 6.10 Bode plot for a transfer function with complex poles: (a) magnitude; (b) phase

313

20

Slope  3 0.2

0.4

1

2

4

40 10

v (rad/sec) (a) 60 Approximation

Phase, ⬔G

100 140 180 220 260 300 0.1

0.2

0.4

1

2

4

10

v (rad/sec) (b)

EXAMPLE 6.5

Bode Plot for Complex Poles and Zeros: Satellite with Flexible Appendages As a third example, draw the Bode plots for a system with second-order terms. The transfer function represents a mechanical system with two equal masses coupled with a lightly damped spring. The applied force and position measurement are collocated on the same mass. For the transfer function, the time scale has been chosen so that the resonant frequency of the complex zeros is equal to 1. The transfer function is KG(s) =

0.01(s2 + 0.01s + 1) . s2 [(s2 /4) + 0.02(s/2) + 1]

Solution. Proceeding through the steps, we start with the low-frequency asymptote, 0.01/ω2 . It has a slope of −2 (−40 db per decade) and passes through magnitude = 0.01 at ω = 1, as shown in Fig. 6.11(a). At the break-point frequency of the zero, ω = 1, the slope shifts to zero until the break point of the pole, which is located at ω = 2, when the slope returns to a slope of −2. To interpolate the true curve, we plot

100

40

10

20

0.1

20

0.01

40

0.001

60 0.1

1

10

db

0

1

0.0001

80

v (rad/sec) (a)

20 20 Phase, ⬔G

Figure 6.11 Bode plot for a transfer function with complex poles and zeros: (a) magnitude; (b) phase

Magnitude, G

314 Chapter 6 The Frequency-Response Design Method

60 100 140 180

0.1

1 v (rad/sec)

10

(b)

the point at the zero break point, ω = 1, with a magnitude ratio below the asymptote of 2ζ = 0.01. At the pole break point, the magnitude ratio above the asymptote is 1/2ζ = 1/0.02 = 50. The magnitude curve is a “doublet”of a negative pulse followed by a positive pulse. Figure 6.11(b) shows that the phase curve for this system starts at −180◦ (corresponding to the 1/s2 term), jumps 180◦ to φ = 0 at ω = 1, due to the zeros, and then falls 180◦ back to φ = −180◦ at ω = 2, due to the pole. With such small damping ratios the stepwise approximation is quite good. (We haven’t drawn this on Fig. 6.11(b), because it would not be easily distinguishable from the true phase curve.) Thus, the true composite phase curve is a nearly square pulse between ω = 1 and ω = 2. In actual designs, Bode plots are made with a computer. However, acquiring the ability to determine how Bode plots should behave is a useful skill, because it gives the designer insight into how changes in the compensation parameters will affect the frequency response. This allows the designer to iterate to the best designs more quickly.

6.1 Frequency Response

EXAMPLE 6.6

315

Computer-Aided Bode Plot for Complex Poles and Zeros Repeat Example 6.5 using MATLAB. Solution. To obtain Bode plots using MATLAB, we call the function bode as follows: numG = 0.01*[1 0.01 1]; denG = [0.25 0.01 1 0 0]; sysG = tf(numG,denG); [mag, phase, w] = bode(sysG); loglog(w,squeeze(mag)) semilogx(w,squeeze(phase)) These commands will result in a Bode plot that matches that in Fig. 6.11 very closely. To obtain the magnitude plot in decibels, the last three lines can be replaced with bode(sysG)

Nonminimum-Phase Systems A system with a zero in the RHP undergoes a net change in phase when evaluated for frequency inputs between zero and infinity, which, for an associated magnitude plot, is greater than if all poles and zeros were in the LHP. Such a system is called nonminimum phase. As can be seen from the construction in Fig. WD.3 in Appendix WD, if the zero is in the RHP, then the phase decreases at the zero break point instead of exhibiting the usual phase increase that occurs for an LHP zero. Consider the transfer functions s+1 , G1 (s) = 10 s + 10 s−1 G2 (s) = 10 . s + 10 Both transfer functions have the same magnitude for all frequencies; that is, |G1 ( jω)| = |G2 ( jω)|, as shown in Fig. 6.12(a). But the phases of the two transfer functions are drastically different [Fig. 6.12(b)]. A minimum-phase system (all zeros in the LHP) with a given magnitude curve will produce the smallest net change in the associated phase, as shown in G1 , compared with what the nonminimum-phase system will produce, as shown by the phase of G2 . Hence, G2 is nonminimum phase. The discrepancy between G1 and G2 with regard to the phase change would be greater if two or more zeros of the plant were in the RHP.

6.1.2 Steady-State Errors We saw in Section 4.2 that the steady-state error of a feedback system decreases as the gain of the open-loop transfer function increases. In plotting a composite magnitude

316 Chapter 6 The Frequency-Response Design Method Figure 6.12 Bode plot minimumand nonminimum-phase systems: for (a) magnitude; (b) phase

20

db

Magnitude

10

G1(jv)  G2(jv)

1 0.01

0 0.1

1.0 10 v (rad/sec)

100

1000

100

1000

(a) 180

⬔G2(jv)

Phase

120

60 ⬔G1(jv) 0 0.01

0.1

1.0 10 v (rad/sec) (b)

Position error constant

Velocity error coefficient

curve, we saw in Section 6.1.1 that the open-loop transfer function, at very low frequencies, is approximated by KG( jω) ∼ (6.23) = Ko ( jω)n . Therefore, we can conclude that the larger the value of the magnitude on the lowfrequency asymptote, the lower the steady-state errors will be for the closed-loop system. This relationship is very useful in the design of compensation: Often we want to evaluate several alternate ways to improve stability and to do so we want to be able to see quickly how changes in the compensation will affect the steady-state errors. For a system of the form given by Eq. (6.16)—that is, where n = 0 in Eq. (6.23) (a Type 0 system)—the low-frequency asymptote is a constant and the gain Ko of the open-loop system is equal to the position-error constant Kp . For a unity feedback system with a unit-step input, the Final Value Theorem (Section 3.1.6) was used in Section 4.2.1 to show that the steady-state error is given by 1 ess = . 1 + Kp For a unity-feedback system in which n = −1 in Eq. (6.23), defined to be a Type 1 system in Section 4.2.1, the low-frequency asymptote has a slope of −1. The magnitude of the low-frequency asymptote is related to the gain according to Eq. (6.23);

6.2 Neutral Stability 1000

60

100

40

10

20 0

1

20

0.1 0.01 0.01

db

Ky  10 Magnitude

Figure 6.13 Determination of Kν from the Bode plot for the system 10 KG(s) = s(s+1)

317

0.1

1

40 10

v (rad/sec)

therefore, we can again read the gain, Ko /ω, directly from the Bode magnitude plot. Equation (4.33) tells us that the velocity-error constant Kν = Ko , where, for a unity-feedback system with a unit-ramp input, the steady-state error is 1 ess = . Kν The easiest way of determining the value of Kν in a type 1 system is to read the magnitude of the low-frequency asymptote at ω = 1 rad/sec, because this asymptote is A(ω) = Kν /ω. In some cases the lowest-frequency break point will be below ω = 1 rad/sec; therefore, the asymptote needs to extend to ω = 1 rad/sec in order to read Kν directly. Alternately, we could read the magnitude at any frequency on the low-frequency asymptote and compute it from Kν = ωA(ω).

EXAMPLE 6.7

Computation of Kν As an example of the determination of steady-state errors, a Bode magnitude plot of an open-loop system is shown in Fig. 6.13. Assuming that there is unity feedback as in Fig. 6.4, find the velocity-error constant, Kν . Solution. Because the slope at the low frequencies is −1, we know that the system is Type 1. The extension of the low-frequency asymptote crosses ω = 1 rad/sec at a magnitude of 10. Therefore, Kν = 10 and the steady-state error to a unit ramp for a unity-feedback system would be 0.1. Alternatively, at ω = 0.01 we have |A(ω)| = 1000; therefore, from Eq. (6.23) we have Ko = Kν ∼ = ω|A(ω)| = 0.01(1000) = 10.

6.2 Neutral Stability In the early days of electronic communications, most instruments were judged in terms of their frequency response. It is therefore natural that when the feedback amplifier

318 Chapter 6 The Frequency-Response Design Method was introduced, techniques to determine stability in the presence of feedback were based on this response. Suppose the closed-loop transfer function of a system is known. We can determine the stability of a system by simply inspecting the denominator in factored form (because the factors give the system roots directly) to observe whether the real parts are positive or negative. However, the closed-loop transfer function is usually not known; in fact, the whole purpose behind understanding the root-locus technique is to be able to find the factors of the denominator in the closed-loop transfer function, given only the open-loop transfer function. Another way to determine closed-loop stability is to evaluate the frequency response of the open-loop transfer function KG( jω) and then perform a test on that response. Note that this method also does not require factoring the denominator of the closed-loop transfer function. In this section we will explain the principles of this method. Suppose we have a system defined by Fig. 6.14(a) and whose root locus behaves as shown in Fig. 6.14(b); that is, instability results if K is larger than 2. The neutrally stable points lie on the imaginary axis—that is, where K = 2 and s = j1.0. Furthermore, we saw in Section 5.1 that all points on the locus have the property that |KG(s)| = 1

and

∠G(s) = 180◦ .

At the point of neutral stability we see that these root-locus conditions hold for s = jω, so |KG( jω)| = 1 and ∠G( jω) = 180◦ . (6.24) Thus a Bode plot of a system that is neutrally stable (that is, with K defined such that a closed-loop root falls on the imaginary axis) will satisfy the conditions of Eq. (6.24). Figure 6.15 shows the frequency response for the system whose root locus is plotted in Fig. 6.14 for various values of K. The magnitude response corresponding to K = 2 passes through 1 at the same frequency (ω = 1 rad/sec) at which the phase passes through 180◦ , as predicted by Eq. (6.24). Having determined the point of neutral stability, we turn to a key question: Does increasing the gain increase or decrease the system’s stability? We can see from the root locus in Fig. 6.14(b) that any value of K less than the value at the neutrally stable point will result in a stable system. At the frequency ω where the phase ∠G( jω) = −180◦ (ω = 1 rad/sec), the magnitude |KG( jω)| < 1.0 for stable Figure 6.14 Stability example: (a) system definition; (b) root locus

Im(s) 2 K2

R




K2



1 s(s
1)2

K



Y

2

1

1

Re(s) 1

K2

(a)

K2

2 (b)

6.3 The Nyquist Stability Criterion 40

100 K  10 10

20

K2

0

1

20

0.1 K  0.1

db

Magnitude, KG(jv)

Figure 6.15 Frequency-response magnitude and phase for the system in Fig. 6.14

319

40

0.01

0.001

0.1

1

10

60 100

v (rad/sec)

Phase, ⬔G(jv)

90
80 180 270

Stability condition

35

0.1

1 v (rad/sec)

10

100

values of K and > 1 for unstable values of K. Therefore, we have the following trial stability condition, based on the character of the open-loop frequency response: |KG( jω)| < 1 at ∠G( jω) = −180◦ . (6.25) This stability criterion holds for all systems for which increasing gain leads to instability and |KG( jω)| crosses the magnitude (=1) once, the most common situation. However, there are systems for which an increasing gain can lead from instability to stability; in this case, the stability condition is |KG( jω)| > 1 at ∠G( jω) = −180◦ . (6.26) There are also cases when |KG( jω)| crosses magnitude (=1) more than once. One way to resolve the ambiguity that is usually sufficient is to perform a rough sketch of the root locus. Another, more rigorous, way to resolve the ambiguity is to use the Nyquist stability criterion, the subject of the next section. However, because the Nyquist criterion is fairly complex, it is important while studying it to bear in mind the theme of this section, namely, that for most systems a simple relationship exists between closed-loop stability and the open-loop frequency response.

6.3 The Nyquist Stability Criterion For most systems, as we saw in the previous section, an increasing gain eventually causes instability. In the very early days of feedback control design, this relationship

320 Chapter 6 The Frequency-Response Design Method between gain and stability margins was assumed to be universal. However, designers found occasionally that in the laboratory the relationship reversed itself; that is, the amplifier would become unstable when the gain was decreased. The confusion caused by these conflicting observations motivated Harry Nyquist of the Bell Telephone Laboratories to study the problem in 1932. His study explained the occasional reversals and resulted in a more sophisticated analysis with no loopholes. Not surprisingly, his test has come to be called the Nyquist stability criterion. It is based on a result from complex variable theory known as the argument principle,7 as we briefly explain in this section and in more detail in Appendix WD. The Nyquist stability criterion relates the open-loop frequency response to the number of closed-loop poles of the system in the RHP. Study of the Nyquist criterion will allow you to determine stability from the frequency response of a complex system, perhaps with one or more resonances, where the magnitude curve crosses 1 several times and/or the phase crosses 180◦ several times. It is also very useful in dealing with open-loop, unstable systems, nonminimum-phase systems, and systems with pure delays (transportation lags).

6.3.1 The Argument Principle Consider the transfer function H1 (s) whose poles and zeros are indicated in the s-plane in Fig. 6.16(a). We wish to evaluate H1 for values of s on the clockwise contour C1 . (Hence this is called a contour evaluation.) We choose the test point so for evaluation. The resulting complex quantity has the form H1 (so ) = v = |v|ejα . The value of the argument of H1 (so ) is α = θ1 + θ2 − (φ1 + φ2 ). As s traverses C1 in the clockwise direction starting at so , the angle α of H1 (s) in Fig. 6.16(b) will change (decrease or increase), but it will not undergo a net change of 360◦ as long as there are no poles or zeros within C1 . This is because none of the angles that make up α go through a net revolution. The angles θ1 , θ2 , φ1 , and φ2 increase or decrease as s traverses around C1 , but they return to their original values as s returns to so without rotating through 360◦ . This means that the plot of H1 (s) [Fig. 6.16(b)] will not encircle the origin. This conclusion follows from the fact that α is the sum of the angles indicated in Fig. 6.16(a), so the only way that α can be changed by 360◦ after s executes one full traverse of C1 is for C1 to contain a pole or zero. Now consider the function H2 (s), whose pole–zero pattern is shown in Fig. 6.16(c). Note that it has a singularity (pole) within C1 . Again, we start at the test point so . As s traverses in the clockwise direction around C1 , the contributions from the angles θ1 , θ2 , and φ1 change, but they return to their original values as soon as s returns to so . In contrast, φ2 , the angle from the pole within C1 , undergoes a net change of −360◦ after one full traverse of C1 . Therefore, the argument of H2 (s) undergoes the same change, causing H2 to encircle the origin in the counterclockwise direction, as shown in Fig. 6.16(d). The behavior would be similar if the contour C1 had enclosed a zero instead of a pole. The mapping of C1 would again enclose the origin once in the H2 (s)-plane, except it would do so in the clockwise direction. 7 Sometimes referred to as “Cauchy’s Principle of the Argument.”

6.3 The Nyquist Stability Criterion Figure 6.16 Contour evaluations: (a) s-plane plot of poles and zeros of H1 (s) and the contour C1 ; (b) H1 (s) for s on C1 ; (c) s-plane plot of poles and zeros of H2 (s) and the contour C1 ; (d) H2 (s) for s on C1

Im(s)

321

Im[H1(s)] H1(s) C1

u1

ˆs0

s0 f1

a

f2

Re[H1(s)]

Re(s)

u2

(a)

(b)

Im[H2(s)]

Im(s)

ˆs0 u1

H2(s)

C1 s0 f1

f2

a Re(s)

Re[H2(s)]

u2

(c)

(d)

Thus we have the essence of the argument principle: Argument principle

A contour map of a complex function will encircle the origin Z − P times, where Z is the number of zeros and P is the number of poles of the function inside the contour. For example, if the number of poles and zeros within C1 is the same, the net angles cancel and there will be no net encirclement of the origin.

6.3.2 Application to Control Design To apply the principle to control design, we let the C1 contour in the s-plane encircle the entire RHP, the region in the s-plane where a pole would cause an unstable system (Fig. 6.17). The resulting evaluation of H(s) will encircle the origin only if H(s) has an RHP pole or zero. As stated earlier, what makes all this contour behavior useful is that a contour evaluation of an open-loop KG(s) can be used to determine stability of the closed-loop

322 Chapter 6 The Frequency-Response Design Method Figure 6.17 An s-plane plot of a contour C1 that encircles the entire RHP

Im(s) Contour at infinity

C1 Re(s) C1

Figure 6.18 Block diagram for Y (s)/R(s) = KG(s)/[1 + KG(s)]

R



KG(s)

Y



Im

[KG(s)]sC1

1




0

[1
KG(s)]sC1

Re

Im

0

Re

Figure 6.19 Evaluations of KG(s) and 1 + KG(s): Nyquist plots system. Specifically, for the system in Fig. 6.18, the closed-loop transfer function is Y (s) KG(s) = T (s) = . R(s) 1 + KG(s) Therefore, the closed-loop roots are the solutions of 1 + KG(s) = 0, and we apply the principle of the argument to the function 1+KG(s). If the evaluation contour of this function of s enclosing the entire RHP contains a zero or pole of 1 + KG(s), then the evaluated contour of 1 + KG(s) will encircle the origin. Notice that 1 + KG(s) is simply KG(s) shifted to the right 1 unit, as shown in Fig. 6.19. Therefore, if the plot of 1 + KG(s) encircles the origin, the plot of KG(s) will encircle

6.3 The Nyquist Stability Criterion

Nyquist plot; polar plot

323

−1 on the real axis. Therefore, we can plot the contour evaluation of the open-loop KG(s), examine its encirclements of −1, and draw conclusions about the origin encirclements of the closed-loop function 1 + KG(s). Presentation of the evaluation of KG(s) in this manner is often referred to as a Nyquist plot, or polar plot, because we plot the magnitude of KG(s) versus the angle of KG(s). To determine whether an encirclement is due to a pole or zero, we write 1+KG(s) in terms of poles and zeros of KG(s): 1 + KG(s) = 1 + K

a(s) + Kb(s) b(s) = . a(s) a(s)

(6.27)

Equation (6.27) shows that the poles of 1 + KG(s) are also the poles of G(s). Because it is safe to assume that the poles of G(s) [or factors of a(s)] are known, the (rare) existence of any of these poles in the RHP can be accounted for. Assuming for now that there are no poles of G(s) in the RHP, an encirclement of −1 by KG(s) indicates a zero of 1 + KG(s) in the RHP, and thus an unstable root of the closed-loop system. We can generalize this basic idea by noting that a clockwise contour C1 enclosing a zero of 1+KG(s)—that is, a closed-loop system root—will result in KG(s) encircling the −1 point in a clockwise direction. Likewise, if C1 encloses a pole of 1 + KG(s)— that is, if there is an unstable open-loop pole—there will be a counterclockwise KG(s) encirclement of −1. Furthermore, if two poles or two zeros are in the RHP, KG(s) will encircle −1 twice, and so on. The net number of clockwise encirclements, N, equals the number of zeros (closed-loop system roots) in the RHP, Z, minus the number of open-loop poles in the RHP, P: N = Z − P. This is the key concept of the Nyquist stability criterion. A simplification in the plotting of KG(s) results from the fact that any KG(s) that represents a physical system will have zero response at infinite frequency (i.e., has more poles than zeros). This means that the big arc of C1 corresponding to s at infinity (Fig. 6.17) results in KG(s) being a point of infinitesimally small value near the origin for that portion of C1 . Therefore, we accomplish a complete evaluation of a physical system KG(s) by letting s traverse the imaginary axis from −j∞ to +j∞ (actually, from −jωh to +jωh , where ωh is large enough that |KG( jω)| is much less than 1 for all ω > ωh ). The evaluation of KG(s) from s = 0 to s = j∞ has already been discussed in Section 6.1 under the context of finding the frequency response of KG(s). Because G(−jω) is the complex conjugate of G( jω), we can easily obtain the entire plot of KG(s) by reflecting the 0 ≤ s ≤ +j∞ portion about the real axis, to get the (−j∞ ≤ s < 0) portion. Hence we see that closed-loop stability can be determined in all cases by examination of the frequency response of the open-loop transfer function on a polar plot. In some applications, models of physical systems are simplified so as to eliminate some high-frequency dynamics. The resulting reduced-order transfer function might have an equal number of poles and zeros. In that case the big arc of C1 at infinity needs to be considered. In practice, many systems behave like those discussed in Section 6.2, so you need not carry out a complete evaluation of KG(s) with subsequent inspection of the −1 encirclements; a simple look at the frequency response may suffice to determine stability. However, in the case of a complex system for which the simplistic rules given

324 Chapter 6 The Frequency-Response Design Method in Section 6.2 become ambiguous, you will want to perform the complete analysis, summarized as follows:

Procedure for Determining Nyquist Stability 1. Plot KG(s) for −j∞ ≤ s ≤ +j∞. Do this by first evaluating KG( jω) for ω = 0 to ωh , where ωh is so large that the magnitude of KG( jω) is negligibly small for ω > ωh , then reflecting the image about the real axis and adding it to the preceding image. The magnitude of KG( jω) will be small at high frequencies for any physical system. The Nyquist plot will always be symmetric with respect to the real axis. The plot is normally created by the NYQUIST MATLAB m-file. 2. Evaluate the number of clockwise encirclements of −1, and call that number N. Do this by drawing a straight line in any direction from −1 to ∞. Then count the net number of left-to-right crossings of the straight line by KG(s). If encirclements are in the counterclockwise direction, N is negative. 3. Determine the number of unstable (RHP) poles of G(s), and call that number P. 4. Calculate the number of unstable closed-loop roots Z: Z = N + P.

(6.28)

For stability we wish to have Z = 0; that is, no characteristic equation roots in the RHP. Let us now examine a rigorous application of the procedure for determining stability using Nyquist plots for some examples.

EXAMPLE 6.8

Nyquist Plot for a Second-Order System Determine the stability properties of the system defined in Fig. 6.20. Solution. The root locus of the system in Fig. 6.20 is shown in Fig. 6.21. It shows that the system is stable for all values of K. The magnitude of the frequency response of KG(s) is plotted in Fig. 6.22(a) for K = 1, and the phase is plotted in Fig. 6.22(b); this is the typical Bode method of presenting frequency response and represents the evaluation of G(s) over the interesting range of frequencies. The same information is replotted in Fig. 6.238 in the Nyquist (polar) plot form. Note how the points A, B, C,

Figure 6.20 Control system for Example 6.8

R








K

1 (s
1)2

Y

8 The shape of this Nyquist plot is a cardioid, meaning “heart-shaped,” plane curve. The name was first

used by de Castillon in the Philosophical Transactions of the Royal Society in 1741. The cardioid is also used in optics.

6.3 The Nyquist Stability Criterion Figure 6.21 Root locus of 1 G(s) = (s+1) 2 with respect to K

325

Im(s) 3 2 Root locus

1

3 2 1

1

2

Re(s)

1

Root locus

2

D, and E are mapped from the Bode plot to the Nyquist plot in Fig. 6.23. The arc from G(s) = +1 (ω = 0) to G(s) = 0 (ω = ∞) that lies below the real axis is derived from Fig. 6.22. The portion of the C1 arc at infinity from Fig. 6.17 transforms into G(s) = 0 in Fig. 6.23; therefore, a continuous evaluation of G(s) with s traversing

20

1.0

0 A

B

C 20

0.1 0.01

40

D

0.001

0.1

1 v (rad/sec)

60 100

10

(a)

0

A B

C

90

D 180

0.1

1 v (rad/sec) (b)

10

E 100

db

Magnitude, G(jv)

10.0

Phase, ⬔G(jv)

Figure 6.22 Open-loop Bode plot for 1 G(s) = (s+1) 2

326 Chapter 6 The Frequency-Response Design Method Figure 6.23 Nyquist plot of the evaluation of KG(s) for s = C1 and K = 1

Im[G(s)] G(s) for s  j to 0 0.5j v  1 v  , E 0.5

0.5

D

v0 1.0 A Re[G(s)] B

v1 0.5j C

G(s) for s  0 to
j

C1 is completed by simply reflecting the lower arc about the real axis. This creates the portion of the contour above the real axis and completes the Nyquist (polar) plot. Because the plot does not encircle −1, N = 0. Also, there are no poles of G(s) in the RHP, so P = 0. From Eq. (6.28), we conclude that Z = 0, which indicates there are no unstable roots of the closed-loop system for K = 1. Furthermore, different values of K would simply change the magnitude of the polar plot, but no positive value of K would cause the plot to encircle −1, because the polar plot will always cross the negative real axis when KG(s) = 0. Thus the Nyquist stability criterion confirms what the root locus indicated: the closed-loop system is stable for all K > 0. The MATLAB statements that will produce this Nyquist plot are numG = 1; denG = [1 2 1]; sysG = tf(numG,denG); w=logspace(−2,2); nyquist(sysG,w);

Often the control systems engineer is more interested in determining a range of gains K for which the system is stable than in testing for stability at a specific value of K. To accommodate this requirement, but to avoid drawing multiple Nyquist plots for various values of the gain, the test can be slightly modified. To do so, we scale KG(s) by K and examine G(s) to determine stability for a range of gains K. This is possible because an encirclement of −1 by KG(s) is equivalent to an encirclement of −1/K by G(s). Therefore, instead of having to deal with KG(s), we need only consider G(s), and count the number of the encirclements of the −1/K point. Applying this idea to Example 6.8, we see that the Nyquist plot cannot encircle the −1/K point. For positive K, the −1/K point will move along the negative real axis, so there will not be an encirclement of G(s) for any value of K > 0. (There are also values of K < 0 for which the Nyquist plot shows the system to be stable; specifically, −1 < K < 0. This result may be verified by drawing the 0◦ locus.)

6.3 The Nyquist Stability Criterion

EXAMPLE 6.9

327

Nyquist Plot for a Third-Order System As a second example, consider the system G(s) = 1/s(s + 1)2 for which the closedloop system is defined in Fig. 6.24. Determine its stability properties using the Nyquist criterion.

Figure 6.24 Control system for Example 6.9

R






1 s(s
1)2

K



Y

Solution. This is the same system discussed in Section 6.2. The root locus in Fig. 6.14(b) shows that this system is stable for small values of K but unstable for large values of K. The magnitude and phase of G(s) in Fig. 6.25 are transformed into the Nyquist plot shown in Fig. 6.26. Note how the points A, B, C, D, and E on the Bode plot of Fig. 6.25 map into those on the Nyquist plot of Fig. 6.26. Also note the large arc at infinity that arises from the open-loop pole at s = 0. This pole creates an infinite magnitude of G(s) at ω = 0; in fact, a pole anywhere on the imaginary axis will create an arc at infinity. To correctly determine the number of −1/K point encirclements, we must draw this arc in the proper half-plane: Should it cross the positive real axis, as shown in Fig. 6.26, or the negative one? It is also necessary to assess whether the arc should sweep out 180◦ (as in Fig. 6.26), 360◦ , or 540◦ . A simple artifice suffices to answer these questions. We modify the C1 contour to take a small detour around the pole either to the right (Fig. 6.27) or to the left. It makes A 10

20

1

0 C

0.1

20

D

40

0.01 E

0.001

0.1

10

1 v (rad/sec)

60 100

(a) 90

A B C

180

D 270

E 0.1

1 v (rad/sec) (b)

10

100

db

Magnitude, G

B

Phase, ⬔G

Figure 6.25 Bode plot for G(s) = 1/s(s + 1)2

328 Chapter 6 The Frequency-Response Design Method Figure 6.26 Nyquist plot9 for G(s) = 1/s(s + 1)2

Im[G(s)] v  0

Arc at  all from v  0

v0 v  1 0.5 C 2

1/Ks

D v  , E 1/Kl

Re[G(s)]

B v0

A v  0


no difference to the final stability question which way, but it is more convenient to go to the right because then no poles are introduced within the C1 contour, keeping the value of P equal to 0. Because the phase of G(s) is the negative of the sum of the angles from all of the poles, we see that the evaluation results in a Nyquist plot moving from +90◦ for s just below the pole at s = 0, across the positive real axis to −90◦ for s just above the pole. Had there been two poles at s = 0, the Nyquist plot at infinity would have executed a full 360◦ arc, and so on for three or more poles. Furthermore, for a pole elsewhere on the imaginary axis, a 180◦ clockwise arc would also result but would be oriented differently than the example shown in Fig. 6.26. The Nyquist plot crosses the real axis at ω = 1 with |G| = 0.5, as indicated by the Bode plot. For K > 0, there are two possibilities for the location of −1/K : inside the two loops of the Nyquist plot, or outside the Nyquist contour completely. For large values of K (Kl in Fig. 6.26), −0.5 < −1/Kl < 0 will lie inside the two loops; hence N = 2, and therefore, Z = 2, indicating that there are two unstable roots. This happens for K > 2. For small values of K (Ks in Fig. 6.26), −1/K lies outside the loops; thus N = 0, and all roots are stable. All this information is in agreement with the root locus in Fig. 6.14(b). (When K < 0, −1/K lies on the positive real axis, then N = 1, which means Z = 1 and the system has one unstable root. The 0◦ root locus will verify this result.) 9 The shape of this Nyquist plot is a translated strophoid plane curve, meaning “a belt with a twist.” The

curve was first studied by Barrow in 1670.

6.3 The Nyquist Stability Criterion Figure 6.27 C1 contour enclosing the RHP for the system in Example 6.9

329

Im(s)

C1

Re(s)

For this and many similar systems, we can see that the encirclement criterion reduces to a very simple test for stability based on the open-loop frequency response: The system is stable if |KG( jω)| < 1 when the phase of G( jω) is 180◦ . Note that this relation is identical to the stability criterion given in Eq. (6.25); however, by using the Nyquist criterion, we don’t require the root locus to determine whether |KG( jω)| < 1 or |KG( jω)| > 1. We draw the Nyquist plot using MATLAB, with numG = 1 ; denG = [1 2 1 0]; sysG = tf(numG,denG); nyquist(sysG) axis([−3 3 −3 3]); The axis command scaled the plot so that only points between +3 and −3 on the real and imaginary axes were included. Without manual scaling, the plot would be scaled based on the maximum values computed by MATLAB and the essential features in the vicinity of the −1 region would be lost. For systems that are open-loop unstable, care must be taken because now P = 0 in Eq. (6.28). We shall see that the simple rules from Section 6.2 will need to be revised in this case.

330 Chapter 6 The Frequency-Response Design Method

EXAMPLE 6.10

Nyquist Plot for an Open-Loop Unstable System The third example is defined in Fig. 6.28. Determine its stability properties using the Nyquist criterion.

Figure 6.28 Control system for Example 6.10

R






s
1 s s 1 10

(

K



)

Y

Solution. The root locus for this system is sketched in Fig. 6.29. The open-loop system is unstable because it has a pole in the RHP. The open-loop Bode plot is shown in Fig. 6.30. Note in the Bode that |KG( jω)| behaves exactly the same as if the pole had been in the LHP. However, ∠G( jω) increases by 90◦ instead of the usual decrease at a pole. Any system with a pole in the RHP is unstable; hence it would be impossible to determine its frequency response experimentally because the system would never reach a steady-state sinusoidal response for a sinusoidal input. It is, however, possible to compute the magnitude and phase of the transfer function according to the rules in Section 6.1. The pole in the RHP affects the Nyquist encirclement criterion, because the value of P in Eq. (6.28) is +1. We convert the frequency-response information of Fig. 6.30 into the Nyquist plot (Fig. 6.31) as in the previous examples. As before, the C1 detour around the pole at s = 0 in Fig. 6.32 creates a large arc at infinity in Fig. 6.31. This arc crosses the negative real-axis because of the 180◦ phase contribution of the pole in the RHP. The real-axis crossing occurs at |G(s)| = 1 because in the Bode plot |G(s)| = 1 when ∠G(s) = 180◦ , which happens to be at ω ∼ = 3 rad/sec. The contour shows two different behaviors, depending on the values of K (> 0). For large values of K (K1 in Fig. 6.31), there is one counterclockwise encirclement; hence N = −1. However, because P = 1 from the RHP pole, Z = N + P = 0, so there are no unstable system roots and the system is stable for K > 1. For small values of K (Ks in Fig. 6.31), N = +1 because of the clockwise encirclement and Z = 2, indicating two unstable roots. This happens if K > 1. These results can be verified

Figure 6.29 Root locus for (s+1) G(s) = s(s/10−1)

Im(s) 4 2

8

6

4

2

2 2 4

4

6

8

10 Re(s)

6.3 The Nyquist Stability Criterion 10

20 db

Magnitude, G

Figure 6.30 Bode plot for (s+1) G(s) = s(s/10−1)

0

1

0.1 0.1

1

10

20 100

10

100

v (rad/sec) (a)

Phase, ⬔G

270 220 180 140 90 0.1

1 v (rad/sec) (b)

Figure 6.31 Nyquist plot10 for (s+1) G(s) = s(s/10−1)

Im[G(s)]

v0

v  10  1/Ks 1.1 1

Re[G(s)] v

1/Kl v0 Arc at  all from v  0

10 The shape of this Nyquist plot is a strophoid.

331

332 Chapter 6 The Frequency-Response Design Method Figure 6.32 C1 contour for Example 6.10

Im(s)

C1 180 Re(s)

qualitatively by the root locus in Fig. 6.29. (If K < 0, −1/K is on the positive real-axis so that N = 0 and Z = 1, indicating the system will have one unstable closed-loop pole, which can be verified by a 0◦ root locus.) As with all systems, the stability boundary occurs at |KG( jω)| = 1 for the phase of ∠G( jω) = 180◦ . However, in this case, |KG( jω)| must be greater than 1 to yield the correct number of −1 point encirclements to achieve stability. To draw the Nyquist plot using MATLAB, use the following commands: numG = [1 1]; denG = [0.1 −1 0]; sysG = tf(numG,denG); nyquist(sysG) axis([−3 3 −3 3])

The existence of the RHP pole in Example 6.10 affected the Bode plotting rules of the phase curve and affected the relationship between encirclements and unstable closed-loop roots because P = 1 in Eq. (6.28). But we apply the Nyquist stability criterion without any modifications. The same is true for systems with a RHP zero; that is, a nonminimum-phase zero has no effect on the Nyquist stability criterion, but the Bode plotting rules are affected.

EXAMPLE 6.11

Nyquist Plot Characteristics Find the Nyquist plot for the third-order system s2 + 3 (s + 1)2 and reconcile the plot with the characteristics of G(s). If the G(s) is to be included in a feedback system as shown in Fig. 6.18, then determine whether the system is stable for all positive values of K. Solution. To draw the Nyquist plot using MATLAB, use the following commands: G(s) = K

6.3 The Nyquist Stability Criterion Figure 6.33 Nyquist plot for Example 6.11

333

3

2

Imaginary axis

1

0

−1

−2

−3 −2 −1.5 −1 −0.5 0 0.5 1 Real axis

1.5

2

2.5

3

numG = [1 0 3]; denG = [1 2 1]; sysG = tf(numG,denG); nyquist(sysG) axis([−2 3 −3 3]) The result is shown in Fig. 6.33.11 Note that there are no arcs at infinity for this case due to the lack of any poles at the origin or on the jω axis. Also note that the Nyquist curve associated with the Bode plot (s = +jω) starts at (3, 0), ends at (1, 0), and, therefore, starts and ends with a phase angle of 0◦ . This is as it should be since the numerator and denominator of G(s) are equal order and there are no singularities at the origin. So the Bode plot should start and end with a zero√phase. Also note that the Nyquist plot goes through (0, 0) as s goes through s = +j 3, as it should since the magnitude equals zero when s is at a zero. Furthermore, note that the phase goes from −120◦ as s approaches (0, 0) to +60◦ as s departs from (0, 0). This behavior follows since a Bode plot phase will jump by +180◦ instantaneously as s passes through a zero on the jω axis. The phase initially decreases as the plot leaves the starting point at (3, 0) because the lowest frequency singularity is the pole at s = −1. Changing the gain, K, will increase or decrease the magnitude of the Nyquist plot but it can never cross the negative real axis. Therefore, the closed-loop system will always be stable for positive K. Exercise: Verify this result by making a rough root locus sketch by hand.

11 The shape of this Nyquist plot is a limaçon, a fact pointed out by the third author’s 17-year-old son, who had recently learned about them in his 10th grade trigonometry class. Limaçon means “snail” in French from the Latin “limax,” and was first investigated by Dürer in 1525.

334 Chapter 6 The Frequency-Response Design Method

6.4 Stability Margins

Gain margin

Phase margin

A large fraction of control system designs behave in a pattern roughly similar to that of the system in Section 6.2 and Example 6.9 in Section 6.3; that is, the system is stable for all small gain values and becomes unstable if the gain increases past a certain critical point. Two commonly used quantities that measure the stability margin for such systems are directly related to the stability criterion of Eq. (6.25): gain margin and phase margin. In this section we will define and use these two concepts to study system design. Another measure of stability, originally defined by O. J. M. Smith (1958), combines these two margins into one and gives a better indication of stability for complicated cases. The gain margin (GM) is the factor by which the gain can be raised before instability results. For the typical case, it can be read directly from the Bode plot (for example, see Fig. 6.15) by measuring the vertical distance between the |KG( jω)| curve and the |KG( jω)| = 1 line at the frequency where ∠G( jω) = 180◦ . We see from the figure that when K = 0.1, the system is stable and GM = 20 (or 26 db). When K = 2, the system is neutrally stable with GM = 1 (0 db), while K = 10 results in an unstable system with GM = 0.2 (−14 db). Note that GM is the factor by which the gain K can be raised before instability results; therefore, |GM| < 1 (or |GM| < 0 db) indicates an unstable system. The GM can also be determined from a root locus with respect to K by noting two values of K: (1) at the point where the locus crosses the jω-axis, and (2) at the nominal closed-loop poles. The GM is the ratio of these two values. Another measure that is used to indicate the stability margin in a system is the phase margin (PM). It is the amount by which the phase of G( jω) exceeds −180◦ when |KG( jω)| = 1, which is an alternative way of measuring the degree to which the stability conditions of Eq. (6.25) are met. For the case in Fig. 6.15, we see that PM ∼ = 80◦ for K = 0.1, PM = 0◦ for K = 2, and PM = −35◦ for K = 10. A positive PM is required for stability. Note that the two stability measures, PM and GM, together determine how far the complex quantity G( jω) passes from the −1 point, which is another way of stating the neutral-stability point specified by Eq. (6.24). The stability margins may also be defined in terms of the Nyquist plot. Figure 6.34 shows that GM and PM are measures of how close the Nyquist plot comes to encircling the −1 point. Again we can see that the GM indicates how much the gain can be raised before instability results in a system like the one in Example 6.9. The PM is

Figure 6.34 Nyquist plot for defining GM and PM

Im[KG(s)] 1/GM

PM 1

KG( jv)

Re[KG(s)]

6.4 Stability Margins

the difference between the phase of G( jω) and 180◦ when KG( jω) crosses the circle |KG(s)| = 1; the positive value of PM is assigned to the stable case (i.e., with no Nyquist encirclements). It is easier to determine these margins directly from the Bode plot than from the Nyquist plot. The term crossover frequency, ωc , is often used to refer to the frequency at which the gain is unity, or 0 db. Figure 6.35 shows the same data plotted in Fig. 6.25, but for the case with K = 1. The same values of PM (= 22◦ ) and GM (= 2) may be obtained from the Nyquist plot shown in Fig. 6.26. The real-axis crossing at −0.5 corresponds to a GM of 1/0.5 or 2 and the PM could be computed graphically by measuring the angle of G( jω) as it crosses |G( jω)| = 1 circle. One of the useful aspects of frequency-response design is the ease with which we can evaluate the effects of gain changes. In fact, we can determine the PM from Fig. 6.35 for any value of K without redrawing the magnitude or phase information. We need only indicate on the figure where |KG( jω)| = 1 for selected trial values of K, as has been done with dashed lines in Fig. 6.36. Now we can see that K = 5 yields an unstable PM of −22◦ , while a gain of K = 0.5 yields a PM of +45◦ . Furthermore, if we wish a certain PM (say 70◦ ), we simply read the value of |G( jω)| corresponding

Figure 6.35 GM and PM from the magnitude and phase plots

10

20

Magnitude, G(jv)

5

db

Crossover frequency

335

1/GM  0.5, GM  2 ( 6 db)

2 1

0

0.5 0.2 0.1 0.2

1 2 v (rad/sec)

10

Phase, ⬔G(jv)

90

180

270

PM  22

0.2

1 v (rad/sec)

2

10

20

336 Chapter 6 The Frequency-Response Design Method Figure 6.36 PM versus K from the frequency-response data

20

10  1/K

K  0.2 db

Magnitude, G(jv)

5 KG(jv)  1 for K  0.5

2 1

0

0.5 KG(jv)  1 for K  5 20 10

0.2 0.1 0.2

2 1 v (rad/sec)

90

Phase, ⬔G(jv)

PM  70 180

270

PM 
45 for K  0.5 PM  22 for K5

0.2

1 2 v (rad/sec)

10

to the frequency that would create the desired PM (here ω = 0.2 rad/sec yields 70◦ , where |G( jω)| = 5), and note that the magnitude at this frequency is 1/K . Therefore, a PM of 70◦ will be achieved with K = 0.2. The PM is more commonly used to specify control system performance because it is most closely related to the damping ratio of the system. This can be seen for the open-loop second-order system ωn2 , s(s + 2ζ ωn ) which, with unity feedback, produces the closed-loop system G(s) =

(6.29)

ωn2 . (6.30) s2 + 2ζ ωn s + ωn2 It can be shown that the relationship between the PM and ζ in this system is ⎡ ⎤ T (s) =

⎢ PM = tan−1 ⎣ 

2ζ 1 + 4ζ 4

− 2ζ 2

⎥ ⎦,

(6.31)

6.4 Stability Margins

337

and this function is plotted in Fig. 6.37. Note that the function is approximately a straight line up to about PM = 60◦ . The dashed line shows a straight-line approximation to the function, where PM ζ ∼ . (6.32) = 100 It is clear that the approximation holds only for PM below about 70◦ . Furthermore, Eq. (6.31) is only accurate for the second-order system of Eq. (6.30). In spite of these limitations, Eq. (6.32) is often used as a rule of thumb for relating the closed-loop damping ratio to PM. It is useful as a starting point; however, it is important always to check the actual damping of a design, as well as other aspects of the performance, before calling the design complete. The gain margin for the second-order system [given by Eq. (6.29)] is infinite (GM = ∞), because the phase curve does not cross −180◦ as the frequency increases. This would also be true for any first- or second-order system. Additional data to aid in evaluating a control system based on its PM can be derived from the relationship between the resonant peak Mr and ζ seen in Fig. 6.3. Note that this figure was derived for the same system [Eq. (6.9)] as Eq. (6.30). We can convert the information in Fig. 6.37 into a form relating Mr to the PM. This is depicted in Fig. 6.38, along with the step-response overshoot Mp . Therefore, we see that, given the PM, one can infer information about what the overshoot of the closed-loop step response would be.

1.0 Damping ratio, z

Figure 6.37 Damping ratio versus PM

0.8 0.6 0.4 0.2 0

0 10 20 30 40 50 60 70 80 Phase margin

n

0.90

4

0.80 0.70

Mr

0.60

2

0.50 0.40 0.30

3

Mp

1

0.20 0.10 0

0 0 10 20 30 40 50 60 70 80 90 Phase margin

Mr (overshoot multiple) Frequency-response resonant peak

n

1.00 Mp (overshoot fraction) Transient-response peak magnitude

Figure 6.38 Transient-response overshoot (Mp ) and frequency-response resonant peak (Mr ) versus PM for 2 ω T (s) = (s2 +2ζ ωn s+ω2 )

338 Chapter 6 The Frequency-Response Design Method

Vector margin

Conditionally stable systems

Many engineers think directly in terms of the PM when judging whether a control system is adequately stabilized. In these terms, a PM = 30◦ is often judged to be the lowest adequate value. In addition to testing the stability of a system design using the PM, a designer would typically also be concerned with meeting a speedof-response specification such as bandwidth, as discussed in Section 6.1. In terms of the frequency-response parameters discussed so far, the crossover frequency would best describe a system’s speed of response. This idea will be discussed further in Sections 6.6 and 6.7. In some cases the PM and GM are not helpful indicators of stability. For firstand second-order systems, the phase never crosses the 180◦ line; hence, the GM is always ∞ and not a useful design parameter. For higher-order systems it is possible to have more than one frequency where |KG( jω)| = 1 or where ∠KG( jω) = 180◦ , and the margins as previously defined need clarification. An example of this can be seen in Fig. 10.12, where the magnitude crosses 1 three times. A decision was made to define PM by the first crossing, because the PM at this crossing was the smallest of the three values and thus the most conservative assessment of stability. A Nyquist plot based on the data in Fig. 10.12 would show that the portion of the Nyquist curve closest to the −1 point was the critical indicator of stability, and therefore use of the crossover frequency yielding the minimum value of PM was the logical choice. At best, a designer needs to be judicious when applying the margin definitions described in Fig. 6.34. In fact, the actual stability margin of a system can be rigorously assessed only by examining the Nyquist plot to determine its closest approach to the −1 point. To aid in this analysis, O. J. M. Smith (1958) introduced the vector margin, which he defined to be the distance to the −1 point from the closest approach of the Nyquist plot.12 Figure 6.39 illustrates the idea graphically. Because the vector margin is a single margin parameter, it removes all the ambiguities in assessing stability that come with using GM and PM in combination. In the past it has not been used extensively due to difficulties in computing it. However, with the widespread availability of computer aids, the idea of using the vector margin to describe the degree of stability is much more feasible. There are certain practical examples in which an increase in the gain can make the system stable. As we saw in Chapter 5, these systems are called conditionally

Figure 6.39 Definition of the vector margin on the Nyquist plot

Im[G(s)]

1 Re[G(s)] Vector margin

12 This value is closely related to the use of the sensitivity function for design and the concept of stability robustness, to be discussed in optional Section 6.7.7.

6.4 Stability Margins

339

stable. A representative root-locus plot for such systems is shown in Fig. 6.40. For a point on the root locus, such as A, an increase in the gain would make the system stable by bringing the unstable roots into the LHP. For point B, either a gain increase or decrease could make the system become unstable. Therefore, several gain margins exist that correspond to either gain reduction or gain increase, and the definition of the GM in Fig. 6.34 is not valid. Figure 6.40 Root locus for a conditionally stable system

Im(s) 4 B A 10

8

6

4

2

Re(s)

4

EXAMPLE 6.12

Stability Properties for a Conditionally Stable System Determine the stability properties as a function of the gain K for the system with the open-loop transfer function K(s + 10)2 . s3 Solution. This is a system for which increasing gain causes a transition from instability to stability. The root locus in Fig. 6.41(a)13 shows that the system is unstable KG(s) =

Figure 6.41 System in which increasing gain leads from instability to stability: (a) root locus; (b) Nyquist plot

Im[G(s)]

Im(s) 25

v0

20 15 10

K7 K5

5 40 30

20

10

5

1

Portion of Nyquist plot portrayed by Bode plots 1.4

10

Re[G(s)]

Re(s) v  ៬

10 20 25

(a) 13 The shape of this root locus is the plane curve limaçon.

(b)

340 Chapter 6 The Frequency-Response Design Method for K < 5 and stable for K > 5. The Nyquist plot in Fig. 6.41(b) was drawn for the stable value K = 7. Determination of the margins according to Fig. 6.34 yields PM = +10◦ (stable) and GM = 0.7 (unstable). According to the rules for stability discussed earlier, these two margins yield conflicting signals on the system’s stability. We resolve the conflict by counting the Nyquist encirclements in Fig. 6.41(b). There is one clockwise encirclement and one counterclockwise encirclement of the −1 point. Hence there are no net encirclements, which confirms that the system is stable for K = 7. For systems like this it is best to resort to the root locus and/or Nyquist plot (rather than the Bode plot) to determine stability.

EXAMPLE 6.13

Nyquist Plot for a System with Multiple Crossover Frequencies Draw the Nyquist plot for the system 85(s + 1)(s2 + 2s + 43.25) + 2s + 82)(s2 + 2s + 101) 85(s + 1)(s + 1 ± 6.5j) , = 2 s (s + 1 ± 9j)(s + 1 ± 10j)

G(s) =

s2 (s2

and determine the stability margins. Solution. The Nyquist plot (Fig. 6.42) shows that there are three crossover frequencies (ω = 0.75, 9.0, and 10.1 rad/sec) with three corresponding PM values of 37◦ , 80◦ , and 40◦ , respectively. However, the key indicator of stability is the proximity of the Nyquist plot as it approaches the −1 point while crossing the real-axis. In this case, only the GM indicates the poor stability margins of this system. The Bode plot for this system (Fig. 6.43) shows the same three crossings of magnitude = 1 at 0.75, 9.0, and 10.1 rad/sec. The GM value of 1.26 from the Bode plot corresponding to ω = 10.4 rad/sec qualitatively agrees with the GM from the Nyquist plot and would be the most useful and unambiguous margin for this example. In summary, many systems behave roughly like Example 6.9, and for them, the GM and PM are well defined and useful. There are also frequent instances of Figure 6.42 Nyquist plot of the complex system in Example 6.13

Im[G(s)] 1

0.5

2

1

1

2

Re[G(s)]

6.5 Bode’s Gain–Phase Relationship 100

40

10

20 0

Magnitude

1

1/GM  0.79 GM  1.26

0.1

20

0.01

40

0.001

60 80

0.0001 0.00001 0.1 0.2 0.4

db

Figure 6.43 Bode plot of the system in Example 6.13

341

100 1 2 4 6 10 20 40 60 100 v (rad/sec) (a)

0 50 Phase

100 150 180 200

PM  37

250 300 0.1 0.2 0.4

1

2 4 6 10 20 40 60 100 v (rad/sec) (b)

more complicated systems with multiple magnitude 1 crossovers or unstable openloop systems for which the stability criteria defined by Fig. 6.34 are ambiguous or incorrect; therefore, we need to verify the GM and PM as previously defined, and/or modify them by reverting back to the Nyquist stability criterion.

6.5 Bode’s Gain–Phase Relationship One of Bode’s important contributions is the following theorem: Bode’s theorem

For any stable minimum-phase system (i.e., one with no RHP zeros or poles), the phase of G( jω) is uniquely related to the magnitude of G( jω). When the slope of |G( jω)| versus ω on a log–log scale persists at a constant value for approximately a decade of frequency, the relationship is particularly simple and is given by (6.33) ∠G( jω) ∼ = n × 90◦ , where n is the slope of |G( jω)| in units of decade of amplitude per decade of frequency. For example, in considering the magnitude curve alone in Fig. 6.44, we see that Eq. (6.33) can be applied to the two frequencies ω1 = 0.1 (where n = −2) and

342 Chapter 6 The Frequency-Response Design Method 100

40

10

20

db

Magnitude, G(jv)

Slope, n  2

0

1 n  1 0.1

0.1

1

10

20

v (rad/sec) v1 Phase, ⬔G(jv)

Figure 6.44 An approximate gain–phase relationship demonstration

v2

90 120 150 180

0.1

1

10

v (rad/sec)

ω2 = 10 (where n = −1), which are a decade removed from the change in slope, to yield the approximate values of phase, −180◦ and −90◦ . The exact phase curve shown in the figure verifies that indeed the approximation is quite good. It also shows that the approximation will degrade if the evaluation is performed at frequencies closer to the change in slope. An exact statement of the Bode gain–phase theorem is  

1 +∞ dM ∠G( jωo ) = W (u) du (in radians), (6.34) π −∞ du where M = log magnitude = ln |G( jω)|, u = normalized frequency = ln(ω/ωo ), ∼ slope n, as defined in Eq. (6.33), dM/du = W (u) = weighting function = ln(coth|u|/2). Figure 6.45 is a plot of the weighting function W (u) and shows how the phase is most dependent on the slope at ωo ; it is also dependent, though to a lesser degree, on slopes at neighboring frequencies. The figure also suggests that the weighting could be approximated by an impulse function centered at ωo . We may approximate the weighting function as π2 W (u) ∼ δ(u), = 2 which is precisely the approximation made to arrive at Eq. (6.33) using the “sifting”property of the impulse function (and conversion from radians to degrees).

6.5 Bode’s Gain–Phase Relationship Figure 6.45 Weighting function in Bode’s gain–phase theorem

343

5 4.5 4 W(u)

3.5 3 2.5 2 1.5 1 0.5 6

4

2

0 u

2

0.1v0

v0

10v0

4

6

In practice, Eq. (6.34) is never used, but Eq. (6.33) is used as a guide to infer stability from |G(ω)| alone. When |KG( jω)| = 1, ∠G( jω) ∼ = −90◦ if n = −1, ∠G( jω) ∼ = −180◦ if n = −2. Crossover frequency

Crossover at −1 slope

For stability we want ∠G( jω) > −180◦ for PM > 0. Therefore, we adjust the |KG( jω)| curve so that it has a slope of −1 at the “crossover” frequency, ωc , (i.e., where |KG( jω)| = 1). If the slope is −1 for a decade above and below the crossover frequency, then PM ∼ = 90◦ ; however, to ensure a reasonable PM, it is usually necessary only to insist that a −1 slope (−20 db per decade) persist for a decade in frequency that is centered at the crossover frequency. We therefore see that there is a very simple design criterion: Adjust the slope of the magnitude curve |KG( jω)| so that it crosses over magnitude 1 with a slope of −1 for a decade around ωc . This criterion will usually be sufficient to provide an acceptable PM, and hence provide adequate system damping. To achieve the desired speed of response, the system gain is adjusted so that the crossover point is at a frequency that will yield the desired bandwidth or speed of response as determined by Eq. (3.60). Recall that the natural frequency ωn , bandwidth, and crossover frequency are all approximately equal, as will be discussed further in Section 6.6.

EXAMPLE 6.14

Use of Simple Design Criterion for Spacecraft Attitude Control For the spacecraft attitude-control problem defined in Fig. 6.46, find a suitable expression for KD(s) that will provide good damping and a bandwidth of approximately 0.2 rad/sec. Also determine the value of the sensitivity function, S, at ω = 0.05 rad/sec in order to evaluate the magnitude of the tracking error for a reference input at that frequency.

344 Chapter 6 The Frequency-Response Design Method


com



Spacecraft G(s) 

KD(s)





60

1000 Slope  2 or 40 db/decade

100

40 20

10

0

1 0.1

1 s2

db

Figure 6.47 Magnitude of the spacecraft’s frequency response

Compensation

Magnitude, G(jv)

Figure 6.46 Spacecraft attitude-control system

0.1

1

10

100 1000

20

v (rad/sec)

Solution. The magnitude of the frequency response of the spacecraft (Fig. 6.47) clearly requires some reshaping, because it has a slope of −2 (or −40 db per decade) everywhere. The simplest compensation to do the job consists of using proportional and derivative terms (a PD compensator), which produces the relation KD(s) = K(TD s + 1).

(6.35)

We will adjust the gain K to produce the desired bandwidth, and adjust break point ω1 = 1/TD to provide the −1 slope at the crossover frequency. The actual design process to achieve the desired specifications is now very simple: We pick a value of K to provide a crossover at 0.2 rad/sec and choose a value of ω1 that is about four times lower than the crossover frequency, so that the slope will be −1 in the vicinity of the crossover. Figure 6.48 shows the steps we take to arrive at the final compensation: 1. Plot |G( jω)|. 2. Modify the plot to include |D( jω)|, with ω1 = 0.05 rad/sec (TD = 20), so that the slope will be ∼ = −1 at ω = 0.2 rad/sec. 3. Determine that |DG| = 100, where the |DG| curve crosses the line ω = 0.2 rad/sec, which is where we want magnitude 1 crossover to be. 4. In order for crossover to be at ω = 0.2 rad/sec, compute K=

1 1 = 0.01. = [|DG|]ω=0.2 100

Therefore, KD(s) = 0.01(20s + 1) will meet the specifications, thus completing the design. If we were to draw the phase curve of KDG, we would find that PM = 75◦ , which is certainly quite adequate. A plot of the closed-loop frequency-response magnitude (Fig. 6.49) shows that, indeed, the crossover frequency and the bandwidth are almost identical in this case. The sensitivity function was defined by Eq. (4.17) and for this problem is

6.5 Bode’s Gain–Phase Relationship Figure 6.48 Compensated open-loop transfer function

10,000

345

80

Slope  2 or 40 db per decade

v1

200

Slope  1 or 20 db per decade

100 KDG  1

G(jv) 20

Figure 6.49 Closed-loop frequency response of T (s) and S(s)

60

D(jv)G(jv)

0.01 0.02

0.1 0.2 v (rad/sec)

K

40 1  0.01 100

1

Magnitude

101

101

T (jω)

100

100

db

Magnitude, DG

1000

db

2000

S(jω) 10

−1

10−2

10−1 ω (rad/sec)

S=

100

10−1

1 . 1 + KDG

It is shown on the graph along with T (s), the output response versus in input command. The frequency response of T confirms that the design achieved the desired bandwidth of 0.2 rad/sec, and it can also be seen that S has the value of 0.2 at ω = 0.05 rad/sec. The step response of the closed-loop system is shown in Fig. 6.50 and its 14% overshoot confirms the adequate damping.

346 Chapter 6 The Frequency-Response Design Method Figure 6.50 Step response for PD compensation

1.2 1.0 0.8 u 0.6 0.4 0.2 0

0

10

20

30

40 50 60 Time (sec)

70

80

90 100

6.6 Closed-Loop Frequency Response The closed-loop bandwidth was defined in Section 6.1 and in Fig. 6.5. Figure 6.3 showed that the natural frequency is always within a factor of two of the bandwidth for a second-order system. In Example 6.14, we designed the compensation so that the crossover frequency was at the desired bandwidth and verified by computation that the bandwidth was identical to the crossover frequency. Generally, the match between the crossover frequency and the bandwidth is not as good as in Example 6.14. We can help establish a more exact correspondence by making a few observations. Consider a system in which |KG( jω)| shows the typical behavior |KG( jω)|  1

for

ω ωc ,

|KG( jω)| 1

for

ω  ωc ,

where ωc is the crossover frequency. The closed-loop frequency-response magnitude is approximated by ⎧   ω ωc ,  KG( jω)  ⎨ 1, ∼ |KG|, ω  ωc . (6.36) |T ( jω)| =  = 1 + KG( jω)  ⎩ In the vicinity of crossover, where |KG( jω)| = 1, |T ( jω)| depends heavily on the PM. A PM of 90◦ means that ∠G( jωc ) = −90◦ , and therefore |T ( jωc )| = 0.707. On the other hand, PM = 45◦ yields |T ( jωc )| = 1.31. The exact evaluation of Eq. (6.36) was used to generate the curves of |T ( jω)| in Fig. 6.51. It shows that the bandwidth for smaller values of PM is typically somewhat greater than ωc , though usually it is less than 2ωc ; thus ωc ≤ ωBW ≤ 2ωc . Another specification related to the closed-loop frequency response is the resonant-peak magnitude Mr , defined in Fig. 6.5. Figures 6.3 and 6.38 show that, for linear systems, Mr is generally related to the damping of the system. In practice, Mr is rarely used; most designers prefer to use the PM to specify the damping of a system, because the imperfections that make systems nonlinear or cause delays usually erode the phase more significantly than the magnitude. As demonstrated in the last example, it is also important in the design to achieve certain error characteristics and these are often evaluated as a function of the input

6.7 Compensation KG(jv) 2.0

T (jv)

PM  22 PM  45

0

1.0 0.7

3 PM  90

vc

db

Magnitude, T(jv) and G(jv)

Figure 6.51 Closed-loop bandwidth with respect to PM

347

0.2

T (jv)  KG(jv) 20

0.1 Bandwidth

2vc

5vc 10vc

v (rad/sec)

or disturbance frequency. In some cases, the primary function of the control system is to regulate the output to a certain constant input in the presence of disturbances. For these situations, the key item of interest for the design would be the closed-loop frequency response of the error with respect to disturbance inputs.

6.7 Compensation As we discussed in Chapters 4 and 5, dynamic elements (or compensation) are typically added to feedback controllers to improve the system’s stability and error characteristics because the process itself cannot be made to have acceptable characteristics with proportional feedback alone. Section 4.3 discussed the basic types of feedback: proportional, derivative, and integral. Section 5.4 discussed three kinds of dynamic compensation: lead compensation, which approximates proportional-derivative (PD) feedback, lag compensation, which approximates proportional-integral (PI) control, and notch compensation, which has special characteristics for dealing with resonances. In this section we discuss these and other kinds of compensation in terms of their frequency-response characteristics. In most cases, the compensation will be implemented in a microprocessor. Techniques for converting the continuous compensation D(s) into a form that can be coded in the computer was briefly discussed in Section 4.4. It will be illustrated further in this section and will be discussed in more detail in Chapter 8. The frequency-response stability analysis to this point has usually considered the closed-loop system to have the characteristic equation 1 + KG(s) = 0. With the introduction of compensation, the closed-loop characteristic equation becomes 1 + KD(s)G(s) = 0, and all the previous discussion in this chapter pertaining to the frequency response of KG(s) applies directly to the compensated case if we apply it to the frequency response of KD(s)G(s). We call this quantity L(s), the “loop gain,” or open-loop transfer function of the system, where L(s) = KD(s)G(s).

348 Chapter 6 The Frequency-Response Design Method Figure 6.52 Frequency response of PD control

10

20

5 2 D(s)

0 db

1 1 v TD

0.5 0.2 0.1

0.1 0.2

1 vT

2

10

20

90 60 ⬔D(s) 30 v 0

0.1 0.2

1 vT

2

1 TD

10

6.7.1 PD Compensation PD compensation

We will start the discussion of compensation design by using the frequency response with PD control. The compensator transfer function, given by D(s) = (TD s + 1),

(6.37)

was shown in Fig. 5.22 to have a stabilizing effect on the root locus of a second-order system. The frequency-response characteristics of Eq. (6.37) are shown in Fig. 6.52. A stabilizing influence is apparent by the increase in phase and the corresponding +1 slope at frequencies above the break point 1/TD . We use this compensation by locating 1/T so that the increased phase occurs in the vicinity of crossover (that is, where D |KD(s)G(s)| = 1), thus increasing the PM. Note that the magnitude of the compensation continues to grow with increasing frequency. This feature is undesirable because it amplifies the high-frequency noise that is typically present in any real system and, as a continuous transfer function, cannot be realized with physical elements. It is also the reason we stated in Section 5.4 that pure derivative compensation gives trouble.

6.7.2 Lead Compensation In order to alleviate the high-frequency amplification of the PD compensation, a firstorder pole is added in the denominator at frequencies substantially higher than the

6.7 Compensation

Lead compensation

349

break point of the PD compensator. Thus the phase increase (or lead) still occurs, but the amplification at high frequencies is limited. The resulting lead compensation has a transfer function of Ts + 1 D(s) = , α < 1, (6.38) αTs + 1 where 1/α is the ratio between the pole/zero break-point frequencies. Figure 6.53 shows the frequency response of this lead compensation. Note that a significant amount of phase lead is still provided, but with much less amplification at high frequencies. A lead compensator is generally used whenever a substantial improvement in damping of the system is required. The phase contributed by the lead compensation in Eq. (6.38) is given by φ = tan−1 (T ω) − tan−1 (αT ω). It can be shown (see Problem 6.44) that the frequency at which the phase is maximum is given by 1 ωmax = √ . (6.39) T α The maximum phase contribution—that is, the peak of the ∠D(s) curve in Fig. 6.53— corresponds to 1−α , (6.40) sin φmax = 1+α or 1 − sin φmax . α= 1 + sin φmax

20

10 5

v

2 D(s)

1 aT

1

0

db

Figure 6.53 Lead-compensation frequency response with 1/α = 10

1 v T

0.5 0.2 0.1

0.1

vmaxT 10 vT

1

100

90 ⬔D(s)

60 30

fmax

0 0.1

1

vmaxT vT

10

100

20

350 Chapter 6 The Frequency-Response Design Method Another way to look at this is the following: The maximum phase occurs at a frequency that lies midway between the two break-point frequencies (sometimes called corner frequencies) on a logarithmic scale, √ 1/ T log ωmax = log √ αT 1 1 = log √ + log √ αT T      1 1 1 = log + log , (6.41) 2 T αT as shown in Fig. 6.53. Alternatively, we may state these results in terms of the pole– zero locations. Rewriting D(s) in the form used for root-locus analysis, we have D(s) =

s+z . s+p

ωmax =

 |z| |p|

Problem 6.44 shows that

(6.42)

(6.43)

and

1 (log |z| + log |p|). (6.44) 2 These results agree with the previous ones if we let z = −1/T and p = −1/αT in Eqs. (6.39) and (6.41). For example, a lead compensator with a zero at s = −2 (T = 0.5) and a pole at s = −10 (αT = 0.1) (and thus α = 15 ) would yield the maximum phase lead at √ ωmax = 2 · 10 = 4.47 rad/sec. log ωmax =

Figure 6.54 Maximum phase increase for lead compensation

90 Maximum phase lead

Lead ratio = α1

The amount of phase lead at the midpoint depends only on α in Eq. (6.40) and is plotted in Fig. 6.54. For α = 1/5, Fig. 6.54 shows that φmax = 40◦ . Note from the figure that we could increase the phase lead up to 90◦ using higher values of the lead ratio, 1/α ; however, Fig. 6.53 shows that increasing values of 1/α also produces higher amplifications at higher frequencies. Thus our task is to select a value of 1/α that is a good compromise between an acceptable PM and an acceptable noise sensitivity at high frequencies. Usually the compromise suggests that a lead compensation should

60

30

0

1

2

4

6 8 10 1

a

20

40 60 100

6.7 Compensation

351

contribute a maximum of 70◦ to the phase. If a greater phase lead is needed, then a double-lead compensation would be suggested, where   Ts + 1 2 D(s) = . αTs + 1 Even if a system had negligible amounts of noise present and the pure derivative compensation of Eq. (6.37) were acceptable, a continuous compensation would look more like Eq. (6.38) than Eq. (6.37) because of the impossibility of building a pure differentiator. No physical system—mechanical or electrical—responds with infinite amplitude at infinite frequencies, so there will be a limit in the frequency range (or bandwidth) for which derivative information (or phase lead) can be provided. This is also true with a digital implementation. Here, the sample rate limits the highfrequency amplification and essentially places a pole in the compensation transfer function.

EXAMPLE 6.15

Lead Compensation for a DC Motor As an example of designing a lead compensator, let us repeat the design of compensation for the DC motor with the transfer function G(s) =

1 s(s + 1)

that was carried out in Section 5.4.1. This also represents the model of a satellite tracking antenna (see Fig. 3.61). This time we wish to obtain a steady-state error of less than 0.1 for a unit-ramp input. Furthermore, we desire an overshoot Mp < 25%. 1. Determine the lead compensation satisfying the specifications. 2. Determine the digital version of the compensation with Ts = 0.05 sec. 3. Compare the step and ramp responses of both implementations. Solution 1. The steady-state error is given by  ess = lim s s→0

 1 R(s), 1 + KD(s)G(s)

(6.45)

where R(s) = 1/s2 for a unit ramp, so Eq. (6.45) reduces to   1 1 . ess = lim = s→0 s + KD(s)[1/(s + 1)] KD(0) Therefore, we find that KD(0), the steady-state gain of the compensation, cannot be less than 10 (Kv ≥ 10) if it is to meet the error criterion, so we pick K = 10. To relate the overshoot requirement to PM, Fig. 6.38 shows that a PM of 45◦ should suffice. The frequency response of KG(s) in Fig. 6.55 shows that the PM = 20◦ if no phase lead is added by compensation. If it were possible to simply add phase without affecting the magnitude, we would need an additional phase of only 25◦ at the KG(s) crossover frequency of ω = 3 rad/sec. However, maintaining the same low-frequency gain and adding a compensator zero would increase the crossover frequency; hence more than a 25◦ phase contribution will

352 Chapter 6 The Frequency-Response Design Method Figure 6.55 Frequency response for lead-compensation design

200 100

40 KD(s)G(s)

10

20

db

Magnitude, L

KG(s) 20

Lead zero 2 1

0 0.1

0.2

1 2 v (rad/sec)

10

Lead pole

Phase (deg) ⬔L

90 120 150 180 PM  20

210 240

0.1

0.2

1 2 v (rad/sec)

53

10

be required from the lead compensation. To be safe, we will design the lead compensator so that it supplies a maximum phase lead of 40◦ . Fig. 6.54 shows that 1/α = 5 will accomplish that goal. We will derive the greatest benefit from the compensation if the maximum phase lead from the compensator occurs at the crossover frequency. With some trial and error, we determine that placing the zero at ω = 2 rad/sec and the pole at ω = 10 rad/sec causes the maximum phase lead to be at the crossover frequency. The compensation, therefore, is s/2 + 1 KD(s) = 10 . s/10 + 1 The frequency-response characteristics of L(s) = KD(s)G(s) in Fig. 6.55 can be seen to yield a PM of 53◦ , which satisfies the design goals. The root locus for this design, originally given as Fig. 5.24, is repeated here as Fig. 6.56, with the root locations marked for K = 10. The locus is not needed for the frequency-response design procedure; it is presented here only for comparison with the root locus design method presented in Chapter 5. The entire process can be expedited by the use of MATLAB’s SISOTOOL routine, which

6.7 Compensation Figure 6.56 Root locus for lead compensation design

353

Im(s) s
1 2 KD(s)  K s
1 10 1 G(s)  s(s
1)

10 8 6 4 2

15

10

5

2

0

5

Re(s)

4

K  10

6 8 10

simultaneously provides the root locus and the Bode plot through an interactive GUI interface. For this example, the MATLAB statements G=tf(1,[1 1 0]); D=tf(10*[1/2 1],[1/10 1]); sisotool(G,D) will provide the plots as shown in Fig. 6.57. It can also be used to generate the Nyquist and time-response plots if desired. 2. To find the discrete equivalent of D(s), we use the trapezoidal rule given by Eq. (4.104). That is, 2 z−1 T z+1 /2 + 1 Dd (z) = 2 s z−1 , (6.46) Ts z+1 /10 + 1 which, with Ts = 0.05 sec, reduces to Dd (z) =

4.2z − 3.8 . z − 0.6

(6.47)

This same result can be obtained by the MATLAB statement sysD = tf([0.5 1],[0.1 1]); sysDd = c2d(sysD, 0.05, ‘tustin’). Because

U(z) = KDd (z), E(z) the discrete control equation that results is u(k + 1) = 0.6u(k) + 10(4.2e(k + 1) − 3.8e(k)). SIMULINK


(6.48)

(6.49)

3. The block diagram of the continuous and discrete versions of D(s) controlling the DC motor is shown in Fig. 6.58. The step responses of the two controllers are plotted together in Fig. 6.59(a) and are reasonably close to one another; however, the discrete controller does exhibit slightly increased overshoot, as is often the case. Both overshoots are less than 25%, and thus

354 Chapter 6 The Frequency-Response Design Method meet the specifications. The ramp responses of the two controllers, shown in Fig. 6.59(b), are essentially identical, and both meet the 0.1 specified error. Figure 6.57 SISOTOOL graphical user interface for Example 6.15

Figure 6.58 SIMULINK
block diagram for transient response of lead-compensation design


Step

Ramp












0.5s
1 0.1s
1

10

Lead control

Slider Kc

4.2z  3.8 z  0.6

10

Discrete lead control

Slider Kd

1 s

1 s
1

Tau 1

Tau 2 Mux

1 s

1 s
1

Tau 1

Tau 2

Mux 1

Output

6.7 Compensation Figure 6.59 Lead-compensation design: (a) step response; (b) ramp response

355

Digital controller 1.2 y 0.8

Continuous controller

0.4 0

0

0.5

1

1.5

2 2.5 3 Time (sec) (a)

4.5

5

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 Time (sec) (b)

2

2

4

Input ramp

1.5 y

3.5

Continuous controller

1 0.5 0

Digital controller 0

The design procedure used in Example 6.15 can be summarized as follows: 1. Determine the low-frequency gain so that the steady-state errors are within specification. 2. Select the combination of lead ratio 1/α and zero values ( 1/T ) that achieves an acceptable PM at crossover. 3. The pole location is then at ( 1/αT ).

Design parameters for lead networks

This design procedure will apply to many cases; however, keep in mind that the specific procedure followed in any particular design may need to be tailored to its particular set of specifications. In Example 6.15 there were two specifications: peak overshoot and steady-state error. We transformed the overshoot specification into a PM, but the steady-state error specification we used directly. No speed-of-response type of specification was given; however, it would have impacted the design in the same way that the steady-state error specification did. The speed of response or bandwidth of a system is directly related to the crossover frequency, as we pointed out earlier in Section 6.6. Figure 6.55 shows that the crossover frequency was ∼ 5 rad/sec. We could have increased it by raising the gain K and increasing the frequency of the lead compensator pole and zero in order to keep the slope of −1 at the crossover frequency. Raising the gain would also have decreased the steady-state error to be better than the specified limit. The gain margin was never introduced into the problem because the stability was adequately specified by the PM alone. Furthermore, the gain margin would not have been useful for this system because the phase never crossed the 180◦ line and the GM was always infinite. In lead-compensation designs there are three primary design parameters: 1. The crossover frequency ωc , which determines bandwidth ωBW , rise time tr , and settling time ts ; 2. The PM, which determines the damping coefficient ζ and the overshoot Mp ; 3. The low-frequency gain, which determines the steady-state error characteristics.

356 Chapter 6 The Frequency-Response Design Method The design problem is to find the best values for the parameters, given the requirements. In essence, lead compensation increases the value of ωc /L(0) (=ωc /Kv for a Type 1 system). That means that, if the low-frequency gain is kept the same, the crossover frequency will increase. Or if the crossover frequency is kept the same, the low-frequency gain will decrease. Keeping this interaction in mind, the designer can assume a fixed value of one of these three design parameters and then adjust the other two iteratively until the specifications are met. One approach is to set the lowfrequency gain to meet the error specifications and add a lead compensator to increase PM at the crossover frequency. An alternative is to pick the crossover frequency to meet a time response specification, then adjust the gain and lead characteristics so that the PM specification is met. A step-by-step procedure is outlined next for these two cases. They apply to a sizable class of problems for which a single lead is sufficient. As with all such design procedures, it provides only a starting point; the designer will typically find it necessary to go through several design iterations in order to meet all the specifications. Design Procedure for Lead Compensation

1. Determine open-loop gain K to satisfy error or bandwidth requirements: (a) to meet error requirement, pick K to satisfy error constants (Kp , Kv , or Ka ) so that ess error specification is met, or alternatively, (b) to meet bandwidth requirement, pick K so that the open-loop crossover frequency is a factor of two below the desired closed-loop bandwidth. 2. Evaluate the PM of the uncompensated system using the value of K obtained from Step 1. 3. Allow for extra margin (about 10◦ ), and determine the needed phase lead φmax . 4. Determine α from Eq. (6.40) or Fig. 6.54. √ 5. Pick ωmax to be at the crossover frequency; thus the zero is at 1/T = ωmax α √ and the pole is at 1/αT = ωmax / α. 6. Draw the compensated frequency response and check the PM. 7. Iterate on the design. Adjust compensator parameters (poles, zeros, and gain) until all specifications are met. Add an additional lead compensator (that is, a double-lead compensation) if necessary. While these guidelines will not apply to all the systems you will encounter in practice, they do suggest a systematic trial-and-error process to search for a satisfactory compensator that will usually be successful.

EXAMPLE 6.16

Lead Compensator for a Temperature Control System The third-order system KG(s) =

K (s/0.5 + 1)(s + 1)(s/2 + 1)

is representative of a typical temperature control system. Design a lead compensator such that Kp = 9 and the PM is at least 25◦ . Solution. Let us follow the design procedure:

6.7 Compensation

357

1. Given the specification for Kp , we solve for K: Kp = lim KG(s) = K = 9. s→0

2. The Bode plot of the uncompensated system, KG(s), with K = 9 can be created by the MATLAB statements below and is shown in Fig. 6.60 along with the two compensated cases. numG = 9; den2 = conv([2 1],[1 1]); denG = conv(den2,[0.5 1]); sysG = tf(numG,denG); w=logspace(−1,1); [mag,phas] = bode(sysG,w); loglog(w,squeeze(mag),grid; semilogx(w,squeeze(phase),grid; It is difficult to read the PM and crossover frequencies accurately from the Bode plots; therefore, the MATLAB command [GM,PM,Wcg,Wcp] = margin(mag,phas,w) can be invoked. The quantity PM is the phase margin and Wcp is the frequency at which the gain crosses magnitude 1. (GM and Wcg are the open-loop gain margin and the frequency at which the phase crosses 180.) For this example, the output is GM =1.25, PM = 7.12, Wcg = 1.87, Wcp = 1.68, which says that the PM of the uncompensated system is 7◦ and that this occurs at a crossover frequency of 1.7 rad/sec. 3. Allowing for 10◦ of extra margin, we want the lead compensator to contribute 25◦ + 10◦ − 7◦ = 28◦ at the crossover frequency. The extra margin is typically 20 KGD1 KGD2

100 101 1 10

100 v (rad/sec) (a)

0

20 101

KG

50 GD2

100 150 200

GD1

180

250 101

100 7 38 16 G v (rad/sec) (b)

db

Magnitude

101

Phase (deg)

Figure 6.60 Bode plot for the lead-compensation design in Example 6.16

101

358 Chapter 6 The Frequency-Response Design Method required because the lead will increase the crossover frequency from the openloop case, at which point more phase increase will be required. From Fig. 6.54 we see that α = 1/3 will produce approximately 30◦ phase increase midway between the zero and pole. As a first cut, let’s place the zero at 1 rad/sec (T = 1) and the pole at 3 rad/sec (αT = 1/3), thus bracketing the open-loop crossover frequency and preserving the factor of 3 between pole and zero, as indicated by α = 1/3. The lead compensator is   1 s+1 s+1 = . D1 (s) = s/3 + 1 0.333 s + 3 The Bode plot of the system with D1 (s) (Fig. 6.60, middle curve) has a PM of 16◦ . We did not achieve the desired PM of 30◦ , because the lead shifted the crossover frequency from 1.7 rad/sec to 2.3 rad/sec, thus increasing the required phase increase from the lead. The step response of the system with D1 (s) (Fig. 6.61) shows a very oscillatory response, as we might expect from the low PM of 16◦ . We repeat the design with extra phase increase and move the zero location slightly to the right so that the crossover frequency won’t be shifted so much. We choose α = 1/10 with the zero at s = −1.5, so   1 s + 1.5 s/1.5 + 1 D2 (s) = = . s/15 + 1 0.1 s + 15 This compensation produces a PM = 38◦ , and the crossover frequency lowered slightly to 2.2 rad/sec. Figure 6.60 (upper curve) shows the frequency response of the revised design. Figure 6.61 shows a substantial reduction in the oscillations, which you should expect from the higher PM value.

4. 5.

6.

7.

Figure 6.61 Step response for lead-compensation design

1.5 1 y 0.5 0

EXAMPLE 6.17

D2 D1

0

2

4

6

8 10 12 Time (sec)

14

16

18

20

Lead-Compensator Design for a Type 1 Servomechanism System Consider the third-order system 10 . s(s/2.5 + 1)(s/6 + 1) This type of system would result for a DC motor with a lag in the shaft position sensor. Design a lead compensator so that the PM = 45◦ and Kv = 10. Solution. Again, we follow the design procedure given earlier: KG(s) = K

1. As given, KG(s) will yield Kv = 10 if K = 1. Therefore, the Kv requirement is met by K = 1 and the low-frequency gain of the compensation should be 1.

6.7 Compensation

359

2. The Bode plot of the system is shown in Fig. 6.62. The PM of the uncompensated system (lower curve) is approximately −4◦ , and the crossover frequency is at ωc ∼ = 4 rad/sec. 3. Allowing for 5◦ of extra PM, we need PM = 45◦ + 5◦ − (−4◦ ) = 54◦ to be contributed by the lead compensator. 4. From Fig. 6.54 we find that α must be 0.1 to achieve a maximum phase lead of 54◦ . 5. The new gain crossover frequency will be higher than the open-loop value of ωc = 4 rad/sec, so let’s select the pole and zero of the lead compensation to be at 20 and 2 rad/sec, respectively. So the candidate compensator is D1 (s) =

1 s+2 s/2 + 1 = . s/20 + 1 0.1 s + 20

6. The Bode plot of the compensated system (Fig. 6.62, middle curve) shows a PM of 23◦ . Further iteration will show that a single-lead compensator cannot meet the specification because of the high-frequency slope of −3. 7. We need a double-lead compensator in this system. If we try a compensator of the form (s/2 + 1)(s/4 + 1) (s + 2)(s + 4) 1 = , D2 (s) = 2 (0.1) (s + 20)(s + 40) (s/20 + 1)(s/40 + 1) we obtain PM = 46◦ . The Bode plot for this case is shown as the upper curve in Fig. 6.62. 20 KGD1

100

KGD2

1

20

10

102 1 10

100

101 KG

50

200

40 102

v(rad/sec) (a) GD1

100 150

0 db

Magnitude

101

Phase (deg)

Figure 6.62 Bode plot for the lead-compensation design in Example 6.17

GD2 180

250 101

G 100

4

23 101 v(rad/sec) (b)

46

102

Both Examples 6.16 and 6.17 are third order. Example 6.17 was more difficult to design compensation for, because the error requirement, Kv , forced the crossover frequency, ωc , to be so high that a single lead could not provide enough PM.

360 Chapter 6 The Frequency-Response Design Method Figure 6.63 Frequency response of PI control

10K

2K D(s)

K v

1 TI

0.2K 0.1K

0.1 0.2

1

2

10

vTI

0 30 ⬔D(s) 60 v 90

0.1 0.2

1

2

1 TI

10

vTI

6.7.3 PI Compensation PI compensation

In many problems it is important to keep the bandwidth low and also to reduce the steady-state error. For this purpose, a proportional-integral (PI) or lag compensator is useful. From Eq. (4.65), we see that PI control has the transfer function   K 1 D(s) = , (6.50) s+ s TI which results in the frequency-response characteristics shown in Fig. 6.63. The desirable aspect of this compensation is the infinite gain at zero frequency, which reduces the steady-state errors. This is accomplished, however, at the cost of a phase decrease at frequencies lower than the break point at ω = 1/TI . Therefore, 1/TI is usually located at a frequency substantially less than the crossover frequency so that the system’s PM is not affected significantly.

6.7.4 Lag Compensation Lag compensation

As we discussed in Section 5.4, lag compensation approximates PI control. Its transfer function was given by Eq. (5.72) for root-locus design, but for frequency-response design, it is more convenient to write the transfer function of the lag compensation alone in the Bode form Ts + 1 D(s) = α , α > 1, (6.51) αTs + 1

6.7 Compensation

10

a 10

5 D(s)

20

2

v

1

1 aT

0 v

1 T

0.1

1 vT

10

0.1

1 vT

10

db

Figure 6.64 Frequency response of lag compensation with α = 10

361

20

0 30 ⬔D(s) 60 90 120

where α is the ratio between the zero/pole break-point frequencies. The complete controller will almost always include an overall gain K and perhaps other dynamics in addition to the lag compensation. Although Eq. (6.51) looks very similar to the lead compensation in Eq. (6.38), the fact is that α > 1 causes the pole to have a lower break-point frequency than the zero. This relationship produces the low-frequency increase in amplitude and phase decrease (lag) apparent in the frequency-response plot in Fig. 6.64 and gives the compensation the essential feature of integral control— an increased low-frequency gain. The typical objective of lag-compensation design is to provide additional gain of α in the low-frequency range and to leave the system sufficient PM. Of course, phase lag is not a useful effect, and the pole and zero of the lag compensator are selected to be at much lower frequencies than the uncompensated system crossover frequency in order to keep the effect on the PM to a minimum. Thus, the lag compensator increases the open-loop DC gain, thereby improving the steady-state response characteristics, without changing the transient response characteristics significantly. If the pole and zero are relatively close together and near the origin (that is, if the value of T is large), we can increase the low-frequency gain (and thus Kp , Kv , or Ka ) by a factor α without moving the closed-loop poles appreciably. Hence, the transient response remains approximately the same while the steady-state response is improved. We now summarize a step-by-step procedure for lag-compensator design. Design Procedure for Lag Compensation

1. Determine the open-loop gain K that will meet the PM requirement without compensation.

362 Chapter 6 The Frequency-Response Design Method 2. Draw the Bode plot of the uncompensated system with crossover frequency from Step 1, and evaluate the low-frequency gain. 3. Determine α to meet the low-frequency gain error requirement. 4. Choose the corner frequency ω = 1/T (the zero of the lag compensator) to be one octave to one decade below the new crossover frequency ωc . 5. The other corner frequency (the pole location of the lag compensator) is then ω = 1/αT . 6. Iterate on the design. Adjust compensator parameters (poles, zeros, and gain) to meet all the specifications.

EXAMPLE 6.18

Lag-Compensator Design for Temperature Control System Again consider the third-order system of Example 6.16: KG(s) = *

1 0.5 s

+

K

+ 1 (s + 1)

*1

+.

+1

2s

Design a lag compensator so the PM is at least 40◦ and Kp = 9. Solution. We follow the design procedure previously enumerated. 1. From the open-loop plot of KG(s), shown for K = 9 in Fig. 6.60, it can be seen that a PM > 40◦ will be achieved if the crossover frequency ωc
1 rad/sec. This will be the case if K = 3. So we pick K = 3 in order to meet the PM specification. 2. The Bode plot of KG(s) in Fig. 6.65 with K = 3 shows that the PM is ≈ 50◦ and the low-frequency gain is now 3. Exact calculation of the PM using MATLAB’s margin shows that PM = 53◦ . KGD 20 KG

100 101 2 10

0

101

20 101

100 v (rad/sec) (a)

50 100 150 200

KG

KGD 180

250 102

101

53 100 v (rad/sec) (b)

db

Magnitude, L

101

Phase (deg) ⬔L

Figure 6.65 Frequency response of lag-compensation design in Example 6.18

101

6.7 Compensation Figure 6.66 Step response of lag-compensation design in Example 6.18

363

1.2 y 0.8 0.4 0

0

2

4

6

8 10 12 Time (sec)

14

16

18

20

3. The low frequency gain should be raised by a factor of 3, which means the lag compensation needs to have α = 3. 4. We choose the corner frequency for the zero to be approximately a factor of 5 slower than the expected crossover frequency—that is, at 0.2 rad/sec. So, 1/T = 0.2, or T = 5. 1 = 5. We then have the value for the other corner frequency: ω = 1/αT = (3)(5) 1/15 rad/sec. The compensator is thus 5s + 1 . D(s) = 3 15s + 1 The compensated frequency response is also shown in Fig. 6.65. The lowfrequency gain of KD(0)G(0) = 3K = 9, thus Kp = 9 and the PM lowers slightly to 44◦ , which satisfies the specifications. The step response of the system, shown in Fig. 6.66, illustrates the reasonable damping that we would expect from PM = 44◦ . 6. No iteration is required in this case. Note that Examples 6.16 and 6.18 are both for the same plant, and both had the same steady-state error requirement. One was compensated with lead and one was compensated with lag. The result is that the bandwidth of the lead-compensated design is higher than that for the lag-compensated design by approximately a factor of 3. This result can be seen by comparing the crossover frequencies of the two designs. A beneficial effect of lag compensation, an increase in the low-frequency gain for better error characteristics, was just demonstrated in Example 6.18. However, in essence, lag compensation reduces the value of ωc /L(0) (=ωc /Kv for a Type 1 system). That means that, if the crossover frequency is kept the same, the low-frequency gain will increase. Likewise, if the low-frequency gain is kept the same, the crossover frequency will decrease. Therefore, lag compensation could also be interpreted to reduce the crossover frequency and thus obtain a better PM. The procedure for design in this case is partially modified. First, pick the low-frequency gain to meet error requirements, then locate the lag compensation pole and zero in order to provide a crossover frequency with adequate PM. The next example illustrates this design procedure. The end result of the design will be the same no matter what procedure is followed.

EXAMPLE 6.19

Lag Compensation of the DC Motor Repeat the design of the DC motor control in Example 6.15, this time using lag compensation. Fix the low-frequency gain in order to meet the error requirement of Kv = 10; then use the lag compensation to meet the PM requirement of 45◦ .

364 Chapter 6 The Frequency-Response Design Method Figure 6.67 Frequency response of lag-compensation design in Example 6.19

KD(s)G(s) KG(s) Lag pole 40

20 20

10

db

Magnitude, L

100

Lag zero 2 1 0.001

0.01

0.1 0.2

1

2

0 10

v (rad/sec)

Phase, ⬔L

90 120 150 180 210 0.001

PM  50 0.01

0.1 0.2

PM  20 1

2

v (rad/sec)

Solution. The frequency response of the system KG(s), with the required gain of K = 10, is shown in Fig. 6.67. The uncompensated system has a crossover frequency at approximately 3 rad/sec where the PM = 20◦ . The designer’s task is to select the lag compensation break points so that the crossover frequency is lowered and more favorable PM results. To prevent detrimental effects from the compensation phase lag, the pole and zero position values of the compensation need to be substantially lower than the new crossover frequency. One possible choice is shown in Fig. 6.67: The lag zero is at 0.1 rad/sec, and the lag pole is at 0.01 rad/sec. This selection of parameters produces a PM of 50◦ , thus satisfying the specifications. Here the stabilization is achieved by keeping the crossover frequency to a region where G(s) has favorable phase characteristics. The criterion for selecting the pole and zero locations 1/T is to make them low enough to minimize the effects of the phase lag from the compensation at the crossover frequency. Generally, however, the pole and zero are located no lower than necessary, because the additional system root (compare with the root locus of a similar system design in Fig. 5.28) introduced by the lag will be in the same frequency range as the compensation zero and will have some effect on the output response, especially the response to disturbance inputs. The response of the system to a step reference input is shown in Fig. 6.68. It shows no steady-state error to a step input, because this is a Type 1 system. However,

6.7 Compensation Figure 6.68 Step response of lag-compensation design in Example 6.19

365

1.2 y

0.8 0.4 0

0

5

10

15

20 25 30 Time (sec)

35

40

45

50

the introduction of the slow root from the lag compensation has caused the response to require about 25 sec to settle down to the zero steady-state value. The overshoot Mp is somewhat larger than you would expect from the guidelines, based on a second-order system shown in Fig. 6.38 for a PM = 50◦ ; however, the performance is adequate. As we saw previously for a similar situation, Examples 6.15 and 6.19 meet an identical set of specifications for the same plant in very different ways. In the first case the specifications are met with a lead compensation, and a crossover frequency ωc = 5 rad/sec (ωBW ∼ = 6 rad/sec) results. In the second case the same specifications are met with a lag compensation, and ωc ∼ = 0.8 rad/sec (ωBW ∼ = 1 rad/sec) results. Clearly, had there been specifications for rise time or bandwidth, they would have influenced the choice of compensation (lead or lag). Likewise, if the slow settling to the steady-state value was a problem, it might have suggested the use of lead compensation instead of lag. In more realistic systems, dynamic elements usually represent the actuator and sensor as well as the process itself, so it is typically impossible to raise the crossover frequency much beyond the value representing the speed of response of the components being used. Although linear analysis seems to suggest that almost any system can be compensated, in fact, if we attempt to drive a set of components much faster than their natural frequencies, the system will saturate, the linearity assumptions will no longer be valid, and the linear design will represent little more than wishful thinking. With this behavior in mind, we see that simply increasing the gain of a system and adding lead compensators to achieve an adequate PM may not always be possible. It may be preferable to satisfy error requirements by adding a lag network so that the closed-loop bandwidth is kept at a more reasonable frequency.

6.7.5 PID Compensation

PID compensation

For problems that need PM improvement at ωc and low-frequency gain improvement, it is effective to use both derivative and integral control. By combining Eqs. (6.37) and (6.50), we obtain PID control. Its transfer function is    K 1 , (6.52) D(s) = (TD s + 1) s + s TI and its frequency-response characteristics are shown in Fig. 6.69. This form is slightly different from that given by Eq. (4.59); however, the effect of the difference is inconsequential. This compensation is roughly equivalent to combining lead and

366 Chapter 6 The Frequency-Response Design Method Figure 6.69 Frequency response of PID compensation with TI /T = 20 D

10K

D(s)

2K K v

0.1 0.2

1

1 TI

2

v

1 TD

10

100

10

100

vTI

90 60 30 ⬔D(s)

0 30 60 90 0.1 0.2

1

2 vTI

lag compensators in the same design, and so is sometimes referred to as a lead– lag compensator. Hence, it can provide simultaneous improvement in transient and steady-state responses.

EXAMPLE 6.20

PID Compensation Design for Spacecraft Attitude Control A simplified design for spacecraft attitude control was presented in Section 6.5; however, here we have a more realistic situation that includes a sensor lag and a disturbing torque. Figure 6.70 defines the system. Design a PID controller to have zero steadystate error to a constant-disturbance torque, a PM of 65◦ , and as high a bandwidth as is reasonably possible. Also evaluate the pointing errors versus frequency and compare them to the errors that would result if the system is open loop. For a torque disturbance from solar pressure that acts as a sinusoid at the orbital rate (ω = 0.001 rad/sec or ≈100-minute period), determine the fractional improvement by the feedback system. Solution. First, let us take care of the steady-state error. For the spacecraft to be at a steady final value, the total input torque, Td + Tc , must equal zero. Therefore, if Td = 0, then Tc = −Td . The only way this can be true with no error (e = 0) is for D(s) to contain an integral term. Hence, including integral control in the compensation will meet the steady-state requirement. This could also be verified mathematically by use of the Final Value Theorem (see Problem 6.47).

6.7 Compensation Figure 6.70 Block diagram of spacecraft control using PID design, Example 6.20

367

Disturbance torque Td

Compensation com








D(s)


Tc






Spacecraft 0.9 s2



Sensor m

2 s
2

The frequency response of the spacecraft and sensor,   0.9 2 G(s) = 2 , (6.53) s s+2 is shown in Fig. 6.71. The slopes of −2 (that is, −40 db per decade) and −3 (−60 db per decade) show that the system would be unstable for any value of K if no derivative feedback were used. This is clear because of Bode’s gain–phase relationship, which shows that the phase would be −180◦ for the −2 slope and −270◦ for the −3 slope and which would correspond to a PM of 0◦ or −90◦ . Therefore, derivative control is required to bring the slope to −1 at the crossover frequency that was shown in Section 6.5 to be a requirement for stability. The problem now is to pick values for the three parameters in Eq. (6.52)—K, TD , and TI —that will satisfy the specifications. The easiest approach is to work first on the phase so that PM = 65◦ is achieved at a reasonably high frequency. This can be accomplished primarily by adjusting TD , noting that TI has a minor effect if sufficiently larger than TD . Once the phase is adjusted, we establish the crossover frequency; then we can easily determine the gain K. We examine the phase of the PID controller in Fig. 6.69 to determine what would happen to the compensated spacecraft system, D(s)G(s), as TD is varied. If 1/T ≥ 2 rad/sec, the phase lead from the PID control would simply cancel the sensor D phase lag, and the composite phase would never exceed −180◦ , an unacceptable situation. If 1/TD ≤ 0.01, the composite phase would approach −90◦ for some range of frequencies and would exceed −115◦ for an even wider range of frequencies; the latter threshold would provide a PM of 65◦ . In the compensated phase curve shown in Fig. 6.71, 1/TD = 0.1, which is the largest value of 1/TD that could provide the required PM of 65◦ . The phase would never cross the −115◦ (65◦ PM) line for any 1/T > 0.1. For 1/T = 0.1, the crossover frequency ωc that produces the 65◦ PM D D is 0.5 rad/sec. For a value of 1/TD 0.05, the phase essentially follows the dotted curve in Fig. 6.71, which indicates that the maximum possible ωc is approximately 1 rad/sec and is provided by 1/TD = 0.05. Therefore, 0.05 < 1/TD < 0.1 is the only sensible range for 1/TD ; anything less than 0.05 would provide no significant increase in bandwidth, while anything more than 0.1 could not meet the PM specification. Although the final choice is somewhat arbitrary, we have chosen 1/TD = 0.1 for our final design. Our choice for 1/TI is a factor of 20 lower than 1/TD ; that is, 1/TI = 0.005. A factor less than 20 would negatively impact the phase at crossover, thus lowering the PM.

368 Chapter 6 The Frequency-Response Design Method Figure 6.71 Compensation for PID design in Example 6.20

vc  0.5 rad/sec 1000

Magnitude, L

1  0.005 TI

40 db

100

60

20 10

D(s)G(s) (K  1)

1  0.1 TD Slope 

1

{

20

2 3

0

20

0.1 G(s) 0.001

0.01

0.1

40

1 2 5 10

100

D(s)G(s),

1  0.05 TD

v (rad/sec)

90

Phase, ⬔L

120

D(s)G(s),

PM 65

150

1  0.1 TD

180 210 ⬔G(s)

240 270

0.001

0.01

0.1

1 2 5 10

100

v (rad/sec)

Furthermore, it is generally desirable to keep the compensated magnitude as large as possible at frequencies below ωc in order to have a faster transient response and smaller errors; maintaining 1/TD and 1/TI at the highest possible frequencies will bring this about. The only remaining task is to determine the proportional part of the PID controller, or K. Unlike the system in Example 6.18, where we selected K in order to meet a steady-state error specification, here we select a value of K that will yield a crossover frequency at the point corresponding to the required PM of 65◦ . The basic procedure for finding K, discussed in Section 6.6, consists of plotting the compensated system amplitude with K = 1, finding the amplitude value at crossover, then setting 1/K equal

6.7 Compensation

369

to that value. Figure 6.71 shows that when K = 1, |D(s)G(s)| = 20 at the desired crossover frequency ωc = 0.5 rad/sec. Therefore, 1 1 = 20, so K = = 0.05. K 20 The compensation equation that satisfies all of the specifications is now complete: 0.05 [(10s + 1)(s + 0.005)]. s It is interesting to note that this system would become unstable if the gain were lowered so that ωc ≤ 0.02 rad/sec, the region in Fig. 6.71 where the phase of the compensated system is less than −180◦ . As mentioned in Section 6.4, this situation is referred to as a conditionally stable system. A root locus with respect to K for this and any conditionally stable system would show the portion of the locus corresponding to very low gains in the RHP. The response of the system for a unit step θcom is shown in Fig. 6.72(a) and exhibits well damped behavior, as should be expected with a 65◦ PM. The response of the system for a step disturbance torque Td = 0.1 N is shown in Fig. 6.72(b). Note that the integral control term does eventually drive the error to zero; however, it is slow due to the presence of a closed-loop pole in the vicinity of the zero at s = −0.005. Recall from the design process that this zero was located in order that the integral term not impact the PM unduly. So if the slow disturbance response is not acceptable, speeding up this pole will decrease the PM and damping of the system. Compromise is often a necessity in control system design! The frequency response of the error characteristics is shown in Fig. 6.73. The top curve is the open-loop error characteristics and the bottom curve is the closed-loop response. The error is attenuated by almost a factor of 106 by the feedback for a disturbance at the orbital rate; there is decreasing error attenuation as the disturbance frequency increases, and there is almost no error attenuation at the system bandwidth of ≈0.5 rad/sec, as you would expect. Note from the design process that the bandwidth was limited by the response characteristics of the sensor, which had a bandwidth of 2 rad/sec. Therefore, the only way to improve the error characteristics would be to D(s) =

1.2 1.0 0.8 u 0.6 0.4 0.2 0

0

5

10

15 20 25 Time (sec)

30

35

40

1.80 1.60 1.40 1.20 1.00 u 0.80 0.60 0.40 0.20 0

0

100 200 300 400 500 600 700 800 900 1000 Time (sec)

(a)

Figure 6.72 Transient response for PID example: (a) step response; (b) step-disturbance response

(b)

370 Chapter 6 The Frequency-Response Design Method 106 105 Open-loop θ error magnitude

Figure 6.73 Frequency response of the error due to a disturbance input, open-loop and closed-loop

104

103 102 Closed-loop 101

100 −3 10

10−2

10−1

100

ω (rad/sec)

increase the bandwidth of the sensor. On the other hand, increasing the bandwidth of the sensor may introduce jitter from the high-frequency sensor noise. Thus we see the classic trade-off dilemma: the designer has to make a judgment as to which feature (low errors due to disturbances or low errors due to sensor noise) is the more important to the overall system performance.

Summary of Compensation Characteristics

1. PD control adds phase lead at all frequencies above the break point. If there is no change in gain on the low-frequency asymptote, PD compensation will increase the crossover frequency and the speed of response. The increase in magnitude of the frequency response at the higher frequencies will increase the system’s sensitivity to noise. 2. Lead compensation adds phase lead at a frequency band between the two break points, which are usually selected to bracket the crossover frequency. If there is no change in gain on the low-frequency asymptote, lead compensation will increase both the crossover frequency and the speed of response over the uncompensated system. 3. PI control increases the frequency-response magnitude at frequencies below the break point, thereby decreasing steady-state errors. It also contributes phase lag below the break point, which must be kept at a low enough frequency to avoid degrading the stability excessively. 4. Lag compensation increases the frequency-response magnitude at frequencies below the two break points, thereby decreasing steady-state errors. Alternatively, with suitable adjustments in K, lag compensation can be used to decrease the frequency-response magnitude at frequencies above the two break points, so that ωc yields an acceptable PM. Lag compensation also contributes phase lag between the two break points, which must be kept at

6.7 Compensation

371

frequencies low enough to keep the phase decrease from degrading the PM excessively. This compensation will typically provide a slower response than using lead compensation.

6.7.6 Design Considerations We have seen in the preceding designs that characteristics of the open-loop Bode plot of the loop gain, L(s) (= KDG), determine performance with respect to steady-state errors, low-frequency errors, and dynamic response. Other properties of feedback, developed in Chapter 4, include reducing the effects of sensor noise and parameter changes on the performance of the system. The consideration of steady-state errors or low-frequency errors due to command inputs and disturbances has been an important design component in the different design methods presented. Design for acceptable errors due to command inputs and disturbances can be thought of as placing a lower bound on the low-frequency gain of the open loop system. Another aspect of the sensitivity issue concerns the highfrequency portion of the system. So far, Chapter 4 and Sections 5.4 and 6.7 have briefly discussed the idea that, to alleviate the effects of sensor noise, the gain of the system at high frequencies must be kept low. In fact, in the development of lead compensation, we added a pole to pure derivative control specifically to reduce the effects of sensor noise at the higher frequencies. It is not unusual for designers to place an extra pole in the compensation, that is, to use the relation D(s) =

Gain stabilization

Phase stabilization

Ts + 1 , (αTs + 1)2

in order to introduce even more attenuation for noise reduction. A second consideration affecting high-frequency gains is that many systems have high-frequency dynamic phenomena, such as mechanical resonances, that could have an impact on the stability of a system. In very-high-performance designs, these highfrequency dynamics are included in the plant model, and a compensator is designed with a specific knowledge of those dynamics. A standard approach to designing for unknown high-frequency dynamics is to keep the high-frequency gain low, just as we did for sensor-noise reduction. The reason for this can be seen from the gain–frequency relationship of a typical system, shown in Fig. 6.74. The only way instability can result from high-frequency dynamics is if an unknown high-frequency resonance causes the magnitude to rise above 1. Conversely, if all unknown high-frequency phenomena are guaranteed to remain below a magnitude of 1, stability can be guaranteed. The likelihood of an unknown resonance in the plant G rising above 1 can be reduced if the nominal high-frequency loop gain (L) is lowered by the addition of extra poles in D(s). When the stability of a system with resonances is assured by tailoring the high-frequency magnitude never to exceed 1, we refer to this process as amplitude or gain stabilization. Of course, if the resonance characteristics are known exactly, a specially tailored compensation, such as one with a notch at the resonant frequency, can be used to change the phase at a specific frequency to avoid encirclements of −1, thus stabilizing the system even though the amplitude does exceed magnitude 1. This method of stabilization is referred to as phase stabilization. A drawback to phase

372 Chapter 6 The Frequency-Response Design Method Figure 6.74 Effect of high-frequency plant uncertainty L(s)

High frequency

1

vc

v

Figure 6.75 Design criteria for low sensitivity

1

1

Steady-state error boundary

Magnitude of L(s)

W2

vc

W1

Sensor noise and plant uncertainty boundary

stabilization is that the resonance information is often not available with adequate precision or varies with time; therefore, the method is more susceptible to errors in the plant model used in the design. Thus, we see that sensitivity to plant uncertainty and sensor noise are both reduced by sufficiently low loop gain at high-frequency. These two aspects of sensitivity—high- and low-frequency behavior—can be depicted graphically, as shown in Fig. 6.75. There is a minimum low-frequency gain allowable for acceptable steady-state and low-frequency error performance and a maximum high-frequency gain allowable for acceptable noise performance and for low probability of instabilities caused by plant-modeling errors. We define the lowfrequency lower bound on the frequency response as W1 and the upper bound as W2−1 , as shown in the figure. Between these two bounds the control engineer must achieve a gain crossover near the required bandwidth; as we have seen, the crossover must occur at a slope of −1 or slightly steeper for good PM and hence damping. For example, if a control system was required to follow a sinusoidal reference input with frequencies from 0 to ω1 with errors no greater than 1%, the function W1 would be 100 from ω = 0 to ω1 . Similar ideas enter into defining possible values for the W2−1 function which would constrain the open-loop gain to be below W2−1 for frequencies above ω2 . These ideas will be discussed further in the following subsections.

v

6.7 Compensation

373


6.7.7 Specifications in Terms of the Sensitivity Function

Figure 6.76 Plot of typical reference spectrum

103

R

102

60

W1

40

101

20

100

0

101 102

101

100 v (rad/sec)

101

db

Sensitivity function

We have seen how the gain and phase margins give useful information about the relative stability of nominal systems and can be used to guide the design of lead and lag compensations. However, the GM and PM are only two numbers and have limitations as guides to the design of realistic control problems. We can express more complete design specifications in the frequency domain if we first give frequency descriptions for the external signals, such as the reference and disturbance, and consider the sensitivity function defined in Section 4.1. For example, we have so far described dynamic performance by the transient response to simple steps and ramps. A more realistic description of the actual complex input signals is to represent them as random processes with corresponding frequency power density spectra. A less sophisticated description, which is adequate for our purposes, is to assume that the signals can be represented as a sum of sinusoids with frequencies in a specified range. For example, we can usually describe the frequency content of the reference input as a sum of sinusoids with relative amplitudes given by a magnitude function |R| such as that plotted in Fig. 6.76, which represents a signal with sinusoidal components having about the same amplitudes up to some value ω1 and very small amplitudes for frequencies above that. With this assumption, the response tracking specification can be expressed by a statement such as “the magnitude of the system error is to be less than the bound eb (a value such as 0.01) for any sinusoid of frequency ωo in the range 0 ≤ ωo ≤ ω1 and of amplitude given by |R( jωo )|.” To express such a performance requirement in terms that can be used in design, we consider again the unity-feedback system drawn in Fig. 6.77. For this system, the error is given by 1 R  S( jω)R, (6.54) E( jω) = 1 + DG where we have used the sensitivity function

20 102

374 Chapter 6 The Frequency-Response Design Method Figure 6.77 Closed-loop block diagram

W R






E

D

U

G









Y











V

1 . (6.55) 1 + DG In addition to being the factor multiplying the system error, the sensitivity function is also the reciprocal of the distance of the Nyquist curve, DG, from the critical point −1. A large value for S indicates a Nyquist plot that comes close to the point of instability. The frequency-based error specification based on Eq. (6.54) can be expressed as |E| = |S| |R| ≤ eb . In order to normalize the problem without needing to define both the spectrum R and the error bound each time, we define the real function of frequency W1 (ω) = |R| /eb and the requirement can be written as S

|S|W1 ≤ 1.

EXAMPLE 6.21

(6.56)

Performance Bound Function A unity-feedback system is to have an error less than 0.005 for all unity amplitude sinusoids below frequency 100 Hertz. Draw the performance frequency function W1 (ω) for this design. Solution. The spectrum, from the problem description, is unity for 0 ≤ ω ≤ 200π rad/sec. Because eb = 0.005, the required function is given by a rectangle of amplitude 1/0.005 = 200 over the given range. The function is plotted in Fig. 6.78. The expression in Eq. (6.56) can be translated to the more familiar Bode plot coordinates and given as a requirement on loop gain by observing that over the frequency range when errors are small the loop gain is large. In that case |S| ≈ 1/|DG|, and the requirement is approximately W1 ≤ 1, |DG| |DG| ≥ W1 .

(6.57)

This requirement can be seen as an extension of the steady-state error requirement from just ω = 0 to the range 0 ≤ ωo ≤ ω1 . In addition to the requirement on dynamic performance, the designer is usually required to design for stability robustness. By this we mean that, while the design is done for a nominal plant transfer function, the actual system is expected to be stable for an entire class of transfer functions that represents the range of changes that are

6.7 Compensation 103

60

102

40

101

20

W1

db

Figure 6.78 Plot of example performance function, W1

100

0

101

102 101

375

20

102

103

40 104

v (rad/sec)

expected to be faced as temperature, age, and other operational and environmental factors vary the plant dynamics from the nominal case. A realistic way to express this uncertainty is to describe the plant transfer function as having a multiplicative uncertainty: (6.58) G( jω) = Go ( jω)[1 + W2 (ω)( jω)]. In Eq. (6.58), the real function W2 is a magnitude function that expresses the size of changes as a function of frequency that the transfer function is expected to experience. In terms of G and Go , the expression is    G − Go  .  W2 =  (6.59) Go  The shape of W2 is almost always very small for low frequencies (we know the model very well there) and increases substantially as we go to higher frequencies, where parasitic parameters come into play and unmodeled structural flexibility is common. A typical shape is sketched in Fig. 6.79. The complex function, ( jω), represents the uncertainty in phase and is restricted only by the constraint 0 ≤ || ≤ 1.

(6.60)

We assume that the nominal design has been done and is stable, so that the Nyquist plot of DGo satisfies the Nyquist stability criterion. In this case, the nominal characteristic equation 1+DGo = 0 is never satisfied for any real frequency. If the system is to have stability robustness, the characteristic equation using the uncertain plant as described by Eq. (6.58) must not go to zero for any real frequency for any value of . The requirement can be written as 1 + DG = 0, 1 + DGo [1 + W2 ] = 0, (1 + DGo )(1 + T W2 ) = 0,

(6.61)

376 Chapter 6 The Frequency-Response Design Method 103

60

102

40

101

20

100

0

db

W2

Figure 6.79 Plot of typical plant uncertainty, W2

101

102 102

20

101

100

101

40 102

v (rad/sec)

where we have defined the complementary sensitivity function as T ( jω)  DGo /(1 + DGo ) = 1 − S.

(6.62)

Because the nominal system is stable, the first term in Eq. (6.61), (1 + DGo ), is never zero. Thus, if Eq. (6.61) is not to be zero for any frequency and any , then it is necessary and sufficient that |T W2 | < 1, which reduces to |T | W2 < 1,

(6.63)

making use of Eq. (6.60). As with the performance specification, for single-input– single-output unity-feedback systems this requirement can be approximated by a more convenient form. Over the range of high frequencies where W2 is non-negligible because there is significant model uncertainty, DGo is small. Therefore we can approximate T ≈ DGo , and the constraint reduces to |DGo | W2 < 1, |DGo | < W1 . 2

(6.64)

The robustness issue is important to design and can affect the high-frequency open-loop frequency response, as discussed above. However, as discussed earlier, it is also important to limit the high-frequency magnitude in order to attenuate noise effects.

DG0 constraint

Figure 6.80 Plot of constraint on |DGo | (= |W2−1 |)

102

40

101

20

100

0

101

102 102

377

db

6.7 Compensation

20

103

104

40 105

v (rad/sec)

EXAMPLE 6.22

Typical Plant Uncertainty The uncertainty in a plant model is described by a function W2 that is zero until ω = 3000, increases linearly from there to a value of 100 at ω = 10,000, and remains at 100 for higher frequencies. Plot the constraint on DGo to meet this requirement. Solution. Where W2 = 0, there is no constraint on the magnitude of loop gain; above ω = 3000, 1/W2 = DGo is a hyperbola from ∞ to 0.01 at ω = 10,000 and remains at 0.01 for ω > 10,000. The bound is sketched in Fig. 6.80. In practice, the magnitude of the loop gain is plotted on log–log (Bode) coordinates, and the constraints of Eqs. (6.57) and (6.64) are included on the same plot. A typical sketch is drawn in Fig. 6.75. The designer is expected to construct a loop gain that will stay above W1 for frequencies below ω1 , cross over the magnitude-1 line (|DG| = 0) in the range ω1 ≤ ω ≤ ω2 , and stay below 1/W2 for frequencies above ω2 .


6.7.8 Limitations on Design in Terms of the Sensitivity Function One of the major contributions of Bode was to derive important limitations on transfer functions that set limits on achievable design specifications. For example, one would like to have the system error kept small for the widest possible range of frequencies and yet have a system that is robustly stable for a very uncertain plant. In terms of the plot in Fig. 6.81, we want W1 and W2 to be very large in their respective frequency ranges and for ω1 to be pushed up close to ω2 . Thus the loop gain is expected to plunge with a large negative slope from being greater than W1 to being less than 1/W in a very short span, while maintaining a good PM to assure stability and good 2

378 Chapter 6 The Frequency-Response Design Method 102

40

W1

20

v2

100

0

v1

101

102 102

db

101 Loop gain constraint

Figure 6.81 Tracking and stability robustness constraints on the Bode plot; an example of impossible constraints

20

100

102

W21

40 104

v (rad/sec)

dynamic performance. The Bode gain–phase formula given earlier shows that this is impossible with a linear controller, by showing that the minimum possible phase is determined by an integral depending on the slope of the magnitude curve. If the slope is constant for a substantial range around ωo , then Eq. (6.34) can be approximated by  π dM  φ(ωo ) ≈ , (6.65) 2 du u=0 where M is the log magnitude and u = log ω/ωo . If, for example, the phase is to be kept above −150◦ to maintain a 30◦ PM, then the magnitude slope near ωo is estimated to be 2
π  dM ≈ −150 du π 180 ≈ −1.667. If we try to make the average slope steeper (more negative) than this, we will lose the PM. From this condition, there developed the design rule that the asymptotes of the Bode plot magnitude, which are restricted to be integral values for rational functions, should be made to cross over the zero-db line at a slope of −1 over a frequency range of about one decade around the crossover frequency, as already discussed in Section 6.5. Modifications to this rule need to be made in particular cases, of course, but the limitation implied by Eq. (6.65) is a hard limit that cannot be avoided. Thus, it is clear that it would be impossible to stabilize the system of Fig. 6.81.

EXAMPLE 6.23

Robustness Constraints If W1 = W2 = 100, and we want PM = 30◦ , what is the minimum ratio of

ω2/ω

1

?

6.7 Compensation

379

Solution. The slope is log W1 − log W12 log ω1 − log ω2

=

2+2 = −1.667. log ωω21

Thus, the log of the ratio is log ω1 /ω2 = −2.40 and ω2 = 251ω1 . An alternative to the standard Bode plot as a design guide can be based on a plot of the sensitivity function as a function of frequency. In this format, Eq. (6.56) requires that |S| < W11 over the range 0 ≤ ω ≤ ω1 for performance, and Eq. (6.64) requires that |S| ≈ 1 over the range ω2 ≤ ω for stability robustness. It should come as no surprise that Bode found a limitation on the possibilities in this case, too. The constraint, extended by Freudenberg and Looze, shows that an integral of the sensitivity function is determined by the presence of poles in the RHP. Suppose the loop gain DGo has np poles, pi , in the RHP and “rolls off” at high frequencies at a slope faster than −1. For rational functions, this means that there is an excess of at least two more finite poles than zeros. Then it can be shown that

∞ 0

Vector margin

ln(|S|) dω = π

np 

Re{pi }.

(6.66)

i=1

If there are no RHP poles, then the integral is zero. This means that if we make the log of the sensitivity function very negative over some frequency band to reduce errors in that band, then, of necessity, ln |S| will be positive over another part of the band, and errors will be amplified there. If there are unstable poles, the situation is worse, because the positive area where sensitivity magnifies the error must exceed the negative area where the error is reduced by the feedback. If the system is minimum phase, then it is, in principle, possible to keep the magnitude of the sensitivity small by spreading the sensitivity increase over all positive frequencies to infinity, but such a design requires an excessive bandwidth and is rarely practical. If a specific bandwidth is imposed, then the sensitivity function is constrained to take on a finite, possibly large, positive value at some point below the bandwidth. As implied by the definition of the vector margin (VM) in Section 6.4 (Fig. 6.39), a large Smax corresponds to a Nyquist plot that comes close to the −1 critical point and a system having a small vector margin, because Smax . (6.67) VM = Smax − 1 If the system is not minimum-phase, the situation is worse. An alternative to Eq. (6.66) is true if there is a nonminimum-phase zero of DGo , a zero in the RHP. Suppose that the zero is located at zo = σo + jωo , where σo > 0. Again, we assume there are np RHP poles at locations pi with conjugate values pi . Now the condition can be expressed as a two-sided weighted integral

  np   pi + zo  σo .  ln(|S|) 2 dω = π ln  σo + (ω − ωo )2 pi − zo  −∞ ∞

i=1

(6.68)

380 Chapter 6 The Frequency-Response Design Method Figure 6.82 Sensitivity function for Example 6.24

101

|S|

100

10−1

10−2 −1 10

100

v (rad/sec)

101

102

In this case, we do not have the “roll-off” restriction, and there is no possibility of spreading the positive area over high frequencies, because the weighting function goes to zero with frequency. The important point about this integral is that if the nonminimum-phase zero is close to a RHP pole, the right side of the integral can be very large, and the excess of positive area is required to be correspondingly large. Based on this result, one expects especially great difficulty meeting both tracking and robustness specifications on sensitivity with a system having RHP poles and zeros close together.

EXAMPLE 6.24

Sensitivity Function for Antenna Compute and plot the sensitivity function for the design of the antenna for which G(s) = 1/s(s + 1) and D(s) = 10(0.5s + 1)/(0.1s + 1). Solution. The sensitivity function for this case is S=

s(s + 1)(s + 10) s + 11s2 + 60s + 100 3

,

(6.69)

and the plot shown in Fig. 6.82 is given by the MATLAB commands numS = [1 11 10 0]; denS = [1 11 60 100]; sysS = tf(numS,denS); [mag,ph,w] = bode(sysS); loglog(w,squeeze(mag)),grid The largest value of S is given by M = max(mag) and is 1.366, from which the vector margin is VM = 3.73.

6.8 Time Delay

381


6.8 Time Delay

Time-delay magnitude

The Laplace transform of a pure time delay is GD (s) = e−sTd and was approximated by a rational function (Padé approximate) in our earlier discussion of root-locus analysis in Chapter 5. Although this same approximation could be used with frequencyresponse methods, an exact analysis of the delay is possible with the Nyquist criterion and Bode plots. The frequency response of the delay is given by the magnitude and phase of e−sTd |s=jω . The magnitude is |GD ( jω)| = |e−jωTd | = 1,

Time-delay phase

for all ω.

(6.70)

This result is expected, because a time delay merely shifts the signal in time and has no effect on its magnitude. The phase is ∠GD ( jω) = −ωTd

(6.71)

in radians, and it grows increasingly negative in proportion to the frequency. This, too, is expected, because a fixed time delay Td becomes a larger fraction or multiple of a sine wave as the period drops, due to increasing frequency. A plot of ∠GD ( jω) is drawn in Fig. 6.83. Note that the phase lag is greater than 270◦ for values of ωTd greater than about 5 rad. This trend implies that it would be virtually impossible to stabilize a system (or to achieve a positive PM) with a crossover frequency greater than ω = 5/Td , and it would be difficult for frequencies greater than ω ∼ = 3/Td . These characteristics essentially place a constraint on the achievable bandwidth of any system with a time delay. (See Problem 6.69 for an illustration of this constraint.) The frequency domain concepts such as the Nyquist criterion apply directly to systems with pure time delay. This means that no approximations (Padé type or otherwise) are needed and the exact effect of time delay can be applied to a Bode plot, as shown in the following example.

EXAMPLE 6.25

Effect of Sampling on Stability Determine the additional phase lag due to the digital sampling in Example 6.15 and reconcile that difference with the observed performance of the continuous and digital

Figure 6.83 Phase lag due to pure time delay

0 90 180 270 ⬔GD(jv) 360 450 540 630 720 0.01

0.1

1

10

vTd (rad)

100 1000

382 Chapter 6 The Frequency-Response Design Method implementations shown in the example. How slowly could you sample if it was necessary to limit the decrease in the PM to less than 20◦ ? Solution. The sample rate in Example 6.15 was selected to be Ts = 0.05 sec. We can see from Fig. 4.22 that the effect of the sampling is to hold the application of the control over one sample period, thus the actual delay varies between zero and one full sample period. Therefore, on the average, the effect of the sampling is to inject a time delay of Ts/2 = 0.05/2 = 0.025 = Td sec. From Eq. (6.71), we see that the phase lag due to this sampling at the crossover frequency of 5 rad/sec, where we measure the PM, is ∠GD = −ωTd = −(5)(0.025) = −0.125 rad = −7◦ . Therefore, the PM will decrease from 45◦ for the continuous implementation to 38◦ for the digital implementation. Fig. 6.59(a) shows that the overshoot, Mp , degraded from 1.2 for the continuous case to ≈1.27 for the digital case, which is predicted by Eq. (6.32) and Fig. 6.38. In order to limit the phase lag to 20◦ at ω = 5 rad/sec, we see from Eq.(6.71) that the maximum tolerable Td = 20/(5 ∗ 57.3) = 0.07 sec, so that the slowest sampling acceptable would be Ts = 0.14 sec. Note, however, that this large decrease in the PM would result in the overshoot increasing from ≈20% to ≈40%. The example illustrates that a time delay, whether introduced by digital sampling or by any other source, has a very severe effect on the achievable bandwidth. Evaluation of the effect using Eq. (6.71) or Fig. 6.83 is simple and straightforward, thus giving a quick analysis of the limitations imposed by any delay in the system. One can also evaluate the effect of a delay using a Nyquist Diagram, and this is shown in Appendix W6.


6.9 Alternative Presentation of Data Before computers were widely available, other ways to present frequency-response data were developed to aid both in understanding design and in easing the designer’s work load. The widespread availability of computers has virtually eliminated the need for these methods. One technique used was the Nichols chart, which we examine in this section because of its place in history. For those interested, we also present the inverse Nyquist method in Appendix W6.

6.9.1 Nichols Chart A rectangular plot of log |G( jω)| versus ∠G( jω) can be drawn by simply transferring the information directly from the separate magnitude and phase portions in a Bode plot; one point on the new curve thus results from a given value of the frequency ω. This means that the new curve is parameterized as a function of frequency. As with the Bode plots, the magnitude information is plotted on a logarithmic scale, while the phase information is plotted on a linear scale. This template was suggested by N. Nichols and is usually referred to as a Nichols chart. The idea of plotting the magnitude of G( jω) versus its phase is similar to the concept of plotting the real and imaginary parts of G( jω), which formed the basis for the Nyquist plots shown in Sections 6.3 and 6.4. However, it is difficult to capture all the pertinent characteristics

6.9 Alternative Presentation of Data

383

of G( jω) on the linear scale of the Nyquist plot. The log scale for magnitude in the Nichols chart alleviates this difficulty, allowing this kind of presentation to be useful for design. For any value of the complex transfer function G( jω), Section 6.6 showed that there is a unique mapping to the unity-feedback closed-loop transfer function G( jω) , 1 + G( jω)

(6.72)

T ( jω) = M(ω)e jα(ω) ,

(6.73)

T ( jω) = or in polar form,

M and N circles

Resonant frequency

EXAMPLE 6.26

where M(ω) is the magnitude of the closed-loop transfer function and α(ω) is the phase of the closed-loop transfer function. Specifically,    G   , (6.74) M= 1 + G G α = tan−1 (N) = ∠ . (6.75) 1+G It can be proven that the contours of constant closed-loop magnitude and phase are circles when G( jω) is presented in the linear Nyquist plot. These circles are referred to as the M and N circles, respectively. The Nichols chart also contains contours of constant closed-loop magnitude and phase based on these relationships, as shown in Fig. 6.84; however, they are no longer circles, because the Nichols charts are semilog plots of magnitude versus linear phase. A designer can therefore graphically determine the bandwidth of a closed-loop system from the plot of the open-loop data on a Nichols chart by noting where the open-loop curve crosses the 0.70 contour of the closed-loop magnitude and determining the frequency of the corresponding data point. Likewise, a designer can determine the resonant peak amplitude Mr by noting the value of the magnitude of the highest closedloop contour tangent to the curve. The frequency associated with the magnitude and phase at the point of tangency is sometimes referred to as the resonant frequency ωr . Similarly, a designer can determine the GM by observing the value of the gain where the Nichols plot crosses the −180◦ line, and the PM by observing the phase where the plot crosses the amplitude 1 line.14 MATLAB provides for easy drawing of a Nichols chart via the nichols m-file.

Nichols Chart for PID Example Determine the bandwidth and resonant peak magnitude of the compensated system whose frequency response is shown in Fig. 6.71. Solution. The magnitude and phase information of the compensated design example seen in Fig. 6.71 is shown on a Nichols chart in Fig. 6.85. When comparing the two figures, it is important to divide the magnitudes in Fig. 6.71 by a factor of 20 in order to obtain |D(s)G(s)| rather than the normalized values used in Fig. 6.71. Because the curve crosses the closed-loop magnitude 0.70 contour at ω = 0.8 rad/sec, we see that 14 James, H. M., N. B. Nichols, and R. S. Phillips (1947).

5

5 0.9

1.00

2

0

5 

5

1.1

1.0

10 9 8 7 6

1

384 Chapter 6 The Frequency-Response Design Method

0

0.9

5

1

1.2

0

0

4

Closedloop magnitude

1.3

0

3

0.80

1.50

0.70

2.0

2

 D(s)G(s)

20

5.0

0.60

1.0 0.9 0.8 0.7 0.6 0.5



0.50

3 0

0.40 0.30

0.4 0.3

0.20 0.2

0.15

Closedloop phase

0.10 0.1 180

160

140

120 100 ⬔D(s)G(s)

80

60

40

Figure 6.84 Nichols chart

the bandwidth of this system is 0.8 rad/sec. Because the largest-magnitude contour touched by the curve is 1.20, we also see that Mr = 1.2. This presentation of data was particularly valuable when a designer had to generate plots and perform calculations by hand. A change in gain, for example, could be evaluated by sliding the curve vertically on transparent paper over a standard Nichols chart as shown in Fig. 6.84. The GM, PM, and bandwidth were then easy to read off the chart, thus allowing evaluations of several values of gain with a minimal amount of effort. With access to computer-aided methods, however, we can now calculate the bandwidth and perform many repetitive evaluations of the gain or any other parameter with a few key strokes. Today the Nichols chart is used primarily as an alternative

5

5 0.9

0

0.9

0

0

1.3

Mr

0

3

0.80

1.50

0.70

2.0

2

 D(s)G(s)

Closedloop magnitude

1

1.2

4

385

5

1.00

2

0 v  0.1 rad/sec

5 

5

1.1

1.0

10 9 8 7 6

1

6.9 Alternative Presentation of Data

5.0

20

0.60 v  0.5 rad/sec

1.0 0.9 0.8 0.7 0.6 0.5



0.50

v  0.8 rad/sec

3

BW v  1 rad/sec

0.4

0

0.40 0.30

0.3

0.20 0.2

0.15

Closedloop phase

v  2 rad/sec

0.10 0.1 180

160

140

120 100 ⬔D(s)G(s)

80

60

40

Figure 6.85 Example plot on the Nichols chart for determining bandwidth and Mr

way to present the information in a Nyquist plot. For complex systems for which the −1 encirclements need to be evaluated, the magnitude log scale of the Nichols chart enables us to examine a wider range of frequencies than a Nyquist plot does, as well as allowing us to read the gain and phase margins directly. Although MATLAB will directly compute PM and GM, the algorithm may lead to suspicious results for very complex cases and the analyst may want to verify the result using the MATLAB nichols m-file so the actual encirclements can be examined and the bases for the PM and GM better understood. An example of the use of a Nichols chart for a complex case is shown in Appendix W6. Another presentation of data is the Inverse Nyquist Diagram, which simplifies the determination of the GM. This is described in more detail in Appendix W6 as well.

386 Chapter 6 The Frequency-Response Design Method

6.10 Historical Perspective As discussed in Chapter 5, engineers before 1960s did not have access to computers to help in their analyses. Therefore, any method that allowed the determination of stability or response characteristics that did not require factoring the characteristic equation was highly useful. The invention of the electronic feedback amplifier by H. S. Black in 1927 at Bell Telephone Laboratories provided extra incentive to develop methods and the development of the frequency response method was the first that enabled design iteration for feedback control design. The development of the feedback amplifier is briefly described in an interesting article based on a talk by Hendrik W. Bode (1960) reproduced in Bellman and Kalaba (1964). With the introduction of electronic amplifiers, long-distance telephoning became possible in the decades following World War I. However, as distances increased, so did the loss of electrical energy; in spite of using larger-diameter wire, increasing numbers of amplifiers were needed to replace the lost energy. Unfortunately, large numbers of amplifiers resulted in much distortion since the small nonlinearity of the vacuum tubes then used in electronic amplifiers was multiplied many times. To solve the problem of reducing distortion, Black proposed the feedback amplifier. As discussed earlier in Chapter 4, the more we wish to reduce errors (or distortion), the higher the feedback needs to be. The loop gain from actuator to plant to sensor to actuator must be made very large. But the designers found that too high a gain produced a squeal and the feedback loop became unstable. In this technology the dynamics were so complex (with differential equations of order 50 being common) that Routh’s criterion, the only way of solving for stability at the time, was not very helpful. So the communications engineers at Bell Telephone Laboratories, familiar with the concept of frequency response and the mathematics of complex variables, turned to complex analysis. In 1932 H. Nyquist published a paper describing how to determine stability from a graphical plot of the open-loop frequency response. Bode then developed his plotting methods in 1938 that made them easy to create without extensive calculations or help from a computer. From the plotting methods and Nyquist’s stability theory there developed an extensive methodology of feedback amplifier design described by Bode (1945) and extensively used still in the design of feedback controls. The reasons for using the method today are primarily to allow for a good design no matter what the unmodeled dynamics are and to expedite the design process, even when carried out with a computer that is fully capable of solving the characteristic equation. After developing the frequency-response design methods prior to World War II, Bode went on to help in electronic fire control devices during the war. The methods that he had developed for feedback amplifiers proved highly applicable to servomechanisms for the effort. Bode characterized this crossover of control system design methods as being a “sort of shotgun marriage.”

SUMMARY •

The frequency-response Bode plot is a graph of the transfer function magnitude in logarithmic scale and the phase in linear scale versus frequency in logarithmic

Summary

387

scale. For a transfer function G(s),

•

• •

•

A = |G( jω)| = |G(s)|s=jω  = {Re[G( jω)]}2 + {Im[G( jω)]}2   −1 Im[G( jω)] = ∠G( jω). φ = tan Re[G( jω)] For a transfer function in Bode form, ( jωτ1 + 1)( jωτ2 + 1) · · · KG(ω) = K0 , ( jωτa + 1)( jωτb + 1) · · · the Bode frequency response can be easily plotted by hand using the rules described in Section 6.1.1. Bode plots can be obtained using computer algorithms (bode in MATLAB), but hand-plotting skills are still extremely helpful. For a second-order system, the peak magnitude of the Bode plot is related to the damping by 1 |G( jω)| = at ω = ωn . 2ζ A method of determining the stability of a closed-loop system based on the frequency response of the system’s open-loop transfer function is the Nyquist stability criterion. Rules for plotting the Nyquist plot are described in Section 6.3. The number of RHP closed-loop roots is given by Z = N + P, where N = number of clockwise encirclements of the −1 point, P = number of open-loop poles in the RHP.

• •

•

•

• •

The Nyquist plot may be obtained using computer algorithms (nyquist in MATLAB). The gain margin (GM) and phase margin (PM) can be determined directly by inspecting the open-loop Bode plot or the Nyquist plot. Also, use of MATLAB’s margin function determines the values directly. For a standard second-order system, the PM is related to the closed-loop damping by Eq. (6.32), ∼ PM . ζ = 100 The bandwidth of the system is a measure of speed of response. For control systems, it is defined as the frequency corresponding to 0.707 (−3 db) in the closed-loop magnitude Bode plot and is approximately given by the crossover frequency ωc , which is the frequency at which the open-loop gain curve crosses magnitude 1. The vector margin is a single-parameter stability margin based on the closest point of the Nyquist plot to the critical point −1/K . For a stable minimum-phase system, Bode’s gain–phase relationship uniquely relates the phase to the gain of the system and is approximated by Eq. (6.33), ∼ n × 90◦ , ∠G( jω) =

388 Chapter 6 The Frequency-Response Design Method Figure 6.86 Typical system

R






E

D

G

Y



•

•

•

•

•

• 

•



•

where n is the slope of |G( jω)| in units of decade of amplitude per decade of frequency. The relationship shows that, in most cases, stability is ensured if the gain plot crosses the magnitude 1 line with a slope of −1. Experimental frequency-response data of the open-loop system can be used directly for analysis and design of a closed-loop control system with no analytical model. For the system shown in Fig. 6.86, the open-loop Bode plot is the frequency response of GD, and the closed-loop frequency response is obtained from T (s) = GD/(1 + GD). The frequency-response characteristics of several types of compensation have been described, and examples of design using these characteristics have been discussed. Design procedures were given for lead and lag compensators in Section 6.7. The examples in that section show the ease of selecting specific values of design variables, a result of using frequency-response methods. A summary was provided at the end of Section 6.7.5. Lead compensation, given by Eq. (6.38), Ts + 1 , α < 1, D(s) = αTs + 1 is a high-pass filter and approximates PD control. It is used whenever substantial improvement in damping of the system is required. It tends to increase the speed of response of a system for a fixed low-frequency gain. Lag compensation, given by Eq. (6.51), Ts + 1 , α > 1, (6.76) D(s) = α αTs + 1 is a low-pass filter and approximates PI control. It is usually used to increase the low-frequency gain of the system so as to improve steady-state response for fixed bandwidth. For a fixed low-frequency gain, it will decrease the speed of response of a system. Tracking-error reduction and disturbance rejection can be specified in terms of the low-frequency gain of the Bode plot. Sensor-noise rejection can be specified in terms of high-frequency attenuation of the Bode plot (see Fig. 6.75). The Nichols plot is an alternate representation of the frequency response as a plot of gain versus phase and is parameterized as a function of frequency. Time delay can be analyzed exactly in a Bode plot or a Nyquist plot.

REVIEW QUESTIONS 1. Why did Bode suggest plotting the magnitude of a frequency response on log–log coordinates? 2.

Define a decibel.

Problems

389

3. What is the transfer function magnitude if the gain is listed as 14 db? 4.

Define gain crossover.

5.

Define phase crossover.

6.

Define phase margin, PM.

7.

Define gain margin, GM.

8. What Bode plot characteristic is the best indicator of the closed-loop step response overshoot? 9. What Bode plot characteristic is the best indicator of the closed-loop step response rise time? 10. What is the principal effect of a lead compensation on Bode plot performance measures? 11. What is the principal effect of a lag compensation on Bode plot performance measures? 12.

How do you find the Kv of a Type 1 system from its Bode plot?

13. Why do we need to know beforehand the number of open-loop unstable poles in order to tell stability from the Nyquist plot? 14. What is the main advantage in control design of counting the encirclements of −1/K of D( jω)G( jω) rather than encirclements of −1 of KD( jω)G( jω)? 15.



Define a conditionally stable feedback system. How can you identify one on a Bode plot?

16. A certain control system is required to follow sinusoids, which may be any frequency in the range 0 ≤ ω ≤ 450 rad/sec and have amplitudes up to 5 units, with (sinusoidal) steady-state error to be never more than 0.01. Sketch (or describe) the corresponding performance function W1 (ω).

PROBLEMS Problems for Section 6.1: Frequency Response 6.1

(a) Show that α0 in Eq. (6.2), with A = Uo and ωo = ω, is   U0 ω  1 = −U0 G(−jω) , α0 = G(s) s − jω s=−jω 2j and

 



U0 ω  1 = U0 G( jω) . α0∗ = G(s) s + jω s=+jω 2j

(b) By assuming the output can be written as y(t) = α0 e−jωt + α0∗ e jωt , derive Eqs. (6.4)–(6.6). 6.2

(a) Calculate the magnitude and phase of G(s) =

1 s + 10

by hand for ω = 1, 2, 5, 10, 20, 50, and 100 rad/sec. (b) Sketch the asymptotes for G(s) according to the Bode plot rules, and compare these with your computed results from part (a).

390 Chapter 6 The Frequency-Response Design Method 6.3

Sketch the asymptotes of the Bode plot magnitude and phase for each of the following open-loop transfer functions. After completing the hand sketches, verify your result using MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. 2000 (a) L(s) = s(s + 200) (b) L(s) =

100 s(0.1s + 1)(0.5s + 1)

(c) L(s) =

1 s(s + 1)(0.02s + 1)

(d) L(s) =

1 (s + 1)2 (s2 + 2s + 4)

(e) L(s) =

10(s + 4) s(s + 1)(s2 + 2s + 5)

(f) L(s) =

1000(s + 0.1) s(s + 1)(s2 + 8s + 64)

(g) L(s) =

(s + 5)(s + 3) s(s + 1)(s2 + s + 4)

4s(s + 10) (s + 100)(4s2 + 5s + 4) s (i) L(s) = (s + 1)(s + 10)(s2 + 2s + 2500)

(h) L(s) =

6.4

6.5

Real poles and zeros. Sketch the asymptotes of the Bode plot magnitude and phase for each of the listed open-loop transfer functions. After completing the hand sketches, verify your result using MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. (a) L(s) =

1 s(s + 1)(s + 5)(s + 10)

(b) L(s) =

(s + 2) s(s + 1)(s + 5)(s + 10)

(c) L(s) =

(s + 2)(s + 6) s(s + 1)(s + 5)(s + 10)

(d) L(s) =

(s + 2)(s + 4) s(s + 1)(s + 5)(s + 10)

Complex poles and zeros. Sketch the asymptotes of the Bode plot magnitude and phase for each of the listed open-loop transfer functions, and approximate the transition at the second-order break point, based on the value of the damping ratio. After completing the hand sketches, verify your result using MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. 1 (a) L(s) = 2 s + 3s + 10 (b) L(s) =

1 s(s2 + 3s + 10)

2 (c) L(s) = (s2 + 2s + 8) s(s + 2s + 10)

Problems

391

2 (d) L(s) = (s 2 + 2s + 12) s(s + 2s + 10) 2 (e) L(s) = (s 2 + 1) s(s + 4) 2 (f) L(s) = (s 2 + 4) s(s + 1)

6.6

Multiple poles at the origin. Sketch the asymptotes of the Bode plot magnitude and phase for each of the listed open-loop transfer functions. After completing the hand sketches, verify your result with MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. (a) L(s) = 2 1 s (s + 8) (b) L(s) = 3 1 s (s + 8) (c) L(s) = 4 1 s (s + 8) (d) L(s) = 2(s + 3) s (s + 8) (e) L(s) = 3(s + 3) s (s + 4) + 1)2 (f) L(s) = (s 3 s (s + 4) 2 (g) L(s) = 3(s + 1) 2 s (s + 10)

6.7

Mixed real and complex poles. Sketch the asymptotes of the Bode plot magnitude and phase for each of the listed open-loop transfer functions. Embellish the asymptote plots with a rough estimate of the transitions for each break point. After completing the hand sketches, verify your result with MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. (a) L(s) =

(s + 2) s(s + 10)(s2 + 2s + 2)

(s + 2) (b) L(s) = 2 s (s + 10)(s2 + 6s + 25) (s + 2)2 (c) L(s) = 2 s (s + 10)(s2 + 6s + 25) 2 (d) L(s) = 2(s + 2)(s 2+ 4s + 68) s (s + 10)(s + 4s + 85) 2 (e) L(s) = 2[(s + 1) + 1] s (s + 2)(s + 3)

392 Chapter 6 The Frequency-Response Design Method 6.8

Right half-plane poles and zeros. Sketch the asymptotes of the Bode plot magnitude and phase for each of the listed open-loop transfer functions. Make sure that the phase asymptotes properly take the RHP singularity into account by sketching the complex plane to see how the ∠L(s) changes as s goes from 0 to +j∞. After completing the hand sketches, verify your result with MATLAB. Turn in your hand sketches and the MATLAB results on the same scales. +2 1 (a) L(s) = ss+ 10 s2 − 1 (the model for a case of magnetic levitation with lead compensation) 1 (b) L(s) = s + 2 (The magnetic levitation system with integral control s(s + 10) (s2 − 1) and lead compensation) (c) L(s) = s −2 1 s (d) L(s) =

s2 + 2s + 1 s(s + 20)2 (s2 − 2s + 2)

(e) L(s) =

(s + 2) s(s − 1)(s + 6)2

1 (s − 1)[(s + 2)2 + 3] A certain system is represented by the asymptotic Bode diagram shown in Fig. 6.87. Find and sketch the response of this system to a unit-step input (assuming zero initial conditions). (f) L(s) =

6.9

Figure 6.87 Magnitude portion of Bode plot for Problem 6.9

10 G

1 0.1

1

10

100

1000

v

6.10 Prove that a magnitude slope of −1 in a Bode plot corresponds to −20 db per decade or −6 db per octave. 6.11 A normalized second-order system with a damping ratio ζ = 0.5 and an additional zero is given by s/a + 1 G(s) = 2 . s +s+1 Use MATLAB to compare the Mp from the step response of the system for a = 0.01, 0.1, 1, 10, and 100 with the Mr from the frequency response of each case. Is there a correlation between Mr and Mp ? 6.12 A normalized second-order system with ζ = 0.5 and an additional pole is given by G(s) =

1 . [(s/p) + 1](s2 + s + 1)

Draw Bode plots with p = 0.01, 0.1, 1, 10, and 100. What conclusions can you draw about the effect of an extra pole on the bandwidth compared with the bandwidth for the second-order system with no extra pole?

Problems

393

6.13 For the closed-loop transfer function ωn2 , T (s) = 2 s + 2ζ ωn s + ωn2 derive the following expression for the bandwidth ωBW of T (s) in terms of ωn and ζ :   ωBW = ωn 1 − 2ζ 2 +

2 + 4ζ 4 − 4ζ 2 .

Assuming that ωn = 1, plot ωBW for 0 ≤ ζ ≤ 1. 6.14 Consider the system whose transfer function is G(s) =

A0 ω0 s . 2 Qs + ω0 s + ω02 Q

This is a model of a tuned circuit with quality factor Q. (a) Compute the magnitude and phase of the transfer function analytically, and plot them for Q = 0.5, 1, 2, and 5 as a function of the normalized frequency ω/ω0 . (b) Define the bandwidth as the distance between the frequencies on either side of ω0 where the magnitude drops to 3 db below its value at ω0 , and show that the bandwidth is given by   1 ω0 BW = . 2π Q (c) What is the relation between Q and ζ ? 6.15 A DC voltmeter schematic is shown in Fig. 6.88. The pointer is damped so that its maximum overshoot to a step input is 10%. (a) What is the undamped natural frequency of the system? (b) What is the damped natural frequency of the system? (c) Plot the frequency response using MATLAB to determine what input frequency will produce the largest magnitude output? (d) Suppose this meter is now used to measure a 1-V AC input with a frequency of 2 rad/sec. What amplitude will the meter indicate after initial transients have died out? What is the phase lag of the output with respect to the input? Use a Bode plot analysis to answer these questions. Use the lsim command in MATLAB to verify your answer in part (d). u

Figure 6.88 Voltmeter schematic I

T k y I ⫽ 40 ⫻ 10⫺6 kg ⭈ m2 k ⫽ 4 ⫻ 10⫺6 kg ⭈ m2/sec2 T ⫽ input torque ⫽ Kmy y ⫽ input voltage Km ⫽ 1 N ⭈ m/V

394 Chapter 6 The Frequency-Response Design Method Problems for Section 6.2: Neutral Stability 6.16 Determine the range of K for which the closed-loop systems (see Fig. 6.18) are stable for each of the cases below by making a Bode plot for K = 1 and imagining the magnitude plot sliding up or down until instability results. Verify your answers by using a very rough sketch of a root-locus plot. + 2) (a) KG(s) = K(s s + 20 (b) KG(s) =

K (s + 10)(s + 1)2

(c) KG(s) = K(s + 10)(s + 1)3 (s + 100)(s + 5) 6.17 Determine the range of K for which each of the listed systems is stable by making a Bode plot for K = 1 and imagining the magnitude plot sliding up or down until instability results. Verify your answers by using a very rough sketch of a root-locus plot. (a) KG(s) = K(s + 1) s(s + 5) + 1) (b) KG(s) = K(s s2 (s + 10) (c) KG(s) =

K (s + 2)(s2 + 9)

+ 1)2 (d) KG(s) = K(s s3 (s + 10)

Problems for Section 6.3: The Nyquist Stability Criterion 6.18

(a) Sketch the Nyquist plot for an open-loop system with transfer function 1/s2 ; that is, sketch  1  , s2 s=C1 where C1 is a contour enclosing the entire RHP, as shown in Fig. 6.17. (Hint: Assume C1 takes a small detour around the poles at s = 0, as shown in Fig. 6.27.) (b) Repeat part (a) for an open-loop system whose transfer function is G(s) = 1/(s2 + ω02 ).

6.19 Sketch the Nyquist plot based on the Bode plots for each of the following systems, and then compare your result with that obtained by using the MATLAB command nyquist: + 2) (a) KG(s) = K(s s + 10 (b) KG(s) =

K (s + 10)(s + 2)2

(c) KG(s) = K(s + 10)(s + 1)3 (s + 100)(s + 2) (d) Using your plots, estimate the range of K for which each system is stable, and qualitatively verify your result by using a rough sketch of a root-locus plot. 6.20 Draw a Nyquist plot for KG(s) =

K(s + 1) , s(s + 3)

(6.77)

Problems

395

choosing the contour to be to the right of the singularity on the jω-axis. Next, using the Nyquist criterion, determine the range of K for which the system is stable. Then redo the Nyquist plot, this time choosing the contour to be to the left of the singularity on the imaginary axis. Again, using the Nyquist criterion, check the range of K for which the system is stable. Are the answers the same? Should they be? 6.21 Draw the Nyquist plot for the system in Fig. 6.89. Using the Nyquist stability criterion, determine the range of K for which the system is stable. Consider both positive and negative values of K.

Figure 6.89 Control system for Problem 6.21

R








K

1 s2  2s  2

Y

1 s1

6.22

(a) For ω = 0.1 to 100 rad/sec, sketch the phase of the minimum-phase system  s + 1  G(s) = s + 10 s=jω and the nonminimum-phase system G(s) = −

 s − 1  , s + 10 s=jω

noting that ∠( jω − 1) decreases with ω rather than increasing. (b) Does an RHP zero affect the relationship between the −1 encirclements on a polar plot and the number of unstable closed-loop roots in Eq. (6.28)? (c) Sketch the phase of the following unstable system for ω = 0.1 to 100 rad/sec:  s + 1  G(s) = . s − 10 s=jω (d) Check the stability of the systems in (a) and (c) using the Nyquist criterion on KG(s). Determine the range of K for which the closed-loop system is stable, and check your results qualitatively by using a rough root-locus sketch. 6.23 Nyquist plots and the classical plane curves: Determine the Nyquist plot, using MATLAB, for the systems given below, with K = 1, and verify that the beginning point and end point for the jω > 0 portion have the correct magnitude and phase: (a) The classical curve called Cayley’s Sextic, discovered by Maclaurin in 1718: KG(s) = K

1 . (s + 1)3

(b) The classical curve called the Cissoid, meaning ivy-shaped: KG(s) = K

1 . s(s + 1)

(c) The classical curve called the Folium of Kepler, studied by Kepler in 1609: KG(s) = K

1 . (s − 1)(s + 1)2

396 Chapter 6 The Frequency-Response Design Method (d) The classical curve called the Folium (not Kepler’s): KG(s) = K

1 . (s − 1)(s + 2)

(e) The classical curve called the Nephroid, meaning kidney-shaped: KG(s) = K

2(s + 1)(s2 − 4s + 1) . (s − 1)3

(f) The classical curve called Nephroid of Freeth, named after the English mathematician T. J. Freeth: (s + 1)(s2 + 3) . KG(s) = K 4(s − 1)3 (g) A shifted Nephroid of Freeth: KG(s) = K

(s2 + 1) . (s − 1)3

Problems for Section 6.4: Stability Margins 6.24 The Nyquist plot for some actual control systems resembles the one shown in Fig. 6.90. What are the gain and phase margin(s) for the system of Fig. 6.90, given that α = 0.4, β = 1.3, and φ = 40◦ . Describe what happens to the stability of the system as the gain goes from zero to a very large value. Sketch what the corresponding root locus must look like for such a system. Also, sketch what the corresponding Bode plots would look like for the system.

Figure 6.90 Nyquist plot for Problem 6.24

Im[G(s)] b

vo

vL

1 f

a

vH

1

Re[G(s)]

v*

6.25 The Bode plot for G(s) = is shown in Fig. 6.91.

100[(s/10) + 1] s[(s/1) − 1][(s/100) + 1]

Problems (a) (b) (c) (d)

Figure 6.91 Bode plot for Problem 6.25

Why does the phase start at −270◦ at the low frequencies? Sketch the Nyquist plot for G(s). Is the closed-loop system shown in Fig. 6.91 stable? Will the system be stable if the gain is lowered by a factor of 100? Make a rough sketch of a root locus for the system, and qualitatively confirm your answer.

1000

60

100

40

10

20

G

397

1

0

0.1

20

0.01

40

0.001

60 0.01

0.1

1 10 v (rad/sec)

100

1000

0.01

0.1

1 10 v (rad/sec)

100

1000

db

90 ⬔G 180 270

6.26 Suppose that in Fig. 6.92, G(s) =

25(s + 1) . s(s + 2)(s2 + 2s + 16)

Use MATLAB’s margin to calculate the PM and GM for G(s) and, on the basis of the Bode plots, conclude which margin would provide more useful information to the control designer for this system.

Figure 6.92 Control system for Problem 6.26

R








G(s)

Y

398 Chapter 6 The Frequency-Response Design Method 6.27 Consider the system given in Fig. 6.93. (a) Use MATLAB to obtain Bode plots for K = 1, and use the plots to estimate the range of K for which the system will be stable. (b) Verify the stable range of K by using margin to determine PM for selected values of K. (c) Use rlocus to determine the values of K at the stability boundaries. (d) Sketch the Nyquist plot of the system, and use it to verify the number of unstable roots for the unstable ranges of K. (e) Using Routh’s criterion, determine the ranges of K for closed-loop stability of this system.

Figure 6.93 Control system for Problem 6.27

R






1 s1

K



Y

s2 (s  1)2  1

6.28 Suppose that in Fig. 6.92, G(s) =

3.2(s + 1) . s(s + 2)(s2 + 0.2s + 16)

Use MATLAB’s margin to calculate the PM and GM for G(s), and comment on whether you think this system will have well-damped closed-loop roots. 6.29 For a given system, show that the ultimate period Pu and the corresponding ultimate gain Ku for the Zeigler–Nichols method can be found by using the following: (a) Nyquist diagram (b) Bode plot (c) Root locus 6.30 If a system has the open-loop transfer function G(s) =

ωn2 s(s + 2ζ ωn )

with unity feedback, then the closed-loop transfer function is given by ωn2 T (s) = 2 . s + 2ζ ωn s + ωn2 Verify the values of the PM shown in Fig. 6.37 for ζ = 0.1, 0.4, and 0.7. 6.31 Consider the unity-feedback system with the open-loop transfer function G(s) =

K . s(s + 1)[(s2 /25) + 0.4(s/5) + 1]

(a) Use MATLAB to draw the Bode plots for G( jω), assuming that K = 1. (b) What gain K is required for a PM of 45◦ ? What is the GM for this value of K? (c) What is Kv when the gain K is set for PM = 45◦ ?

Problems

399

(d) Create a root locus with respect to K, and indicate the roots for a PM of 45◦ . 6.32

For the system depicted in Fig. 6.94(a), the transfer-function blocks are defined by G(s) =

1 (s + 2)2 (s + 4)

and H(s) =

1 . s+1

(a) Using rlocus and rlocfind, determine the value of K at the stability boundary. (b) Using rlocus and rlocfind, determine the value of K that will produce roots with damping corresponding to ζ = 0.707. (c) What is the gain margin of the system if the gain is set to the value determined in part (b)? Answer this question without using any frequency-response methods. (d) Create the Bode plots for the system, and determine the gain margin that results for PM = 65◦ . What damping ratio would you expect for this PM? (e) Sketch a root locus for the system shown in Fig. 6.94(b). How does it differ from the one in part (a)? (f) For the systems in Figs. 6.94(a) and (b), how does the transfer function Y2 (s)/R(s) differ from Y1 (s)/R(s)? Would you expect the step response to r(t) to be different for the two cases?

R






K

H(s)

G(s)

Y1

R






G(s)

K

Y2





H(s) (a)

(b)

Figure 6.94 Block diagram for Problem 6.32: (a) unity feedback; (b) H(s) in feedback 6.33 For the system shown in Fig. 6.95, use Bode and root-locus plots to determine the gain and frequency at which instability occurs. What gain (or gains) gives a PM of 20◦ ? What is the gain margin when PM = 20◦ ?

Figure 6.95 Control system for Problem 6.33

R








K

(s  1)(s  2) s2(s  3)(s2  2s  25)

Y

6.34 A magnetic tape-drive speed-control system is shown in Fig. 6.96. The speed sensor is slow enough that its dynamics must be included. The speed-measurement time constant is τm = 0.5 sec; the reel time constant is τr = J/b = 4 sec, where b = the output shaft damping constant = 1 N·m·sec; and the motor time constant is τ1 = 1 sec. (a) Determine the gain K required to keep the steady-state speed error to less than 7% of the reference-speed setting. (b) Determine the gain and phase margins of the system. Is this a good system design? 6.35 For the system in Fig. 6.97, determine the Nyquist plot and apply the Nyquist criterion (a) to determine the range of values of K (positive and negative) for which the system will be stable, and

400 Chapter 6 The Frequency-Response Design Method Figure 6.96 Magnetic tape-drive speed control

Amplifier and motor vc



K t1s  1






Tape drive Torque

1 Js  b

v

Sensor 1 tm s  1

Figure 6.97 Control system for Problems 6.35, 6.62, and 6.63

R






E

K

F



Y

3 s(s  1)(s  3) Sensor

Yˆ

1

(b) to determine the number of roots in the RHP for those values of K for which the system is unstable. Check your answer by using a rough root-locus sketch. 6.36 For the system shown in Fig. 6.98, determine the Nyquist plot and apply the Nyquist criterion (a) to determine the range of values of K (positive and negative) for which the system will be stable, and (b) to determine the number of roots in the RHP for those values of K for which the system is unstable. Check your answer by using a rough root-locus sketch.

Figure 6.98 Control system for Problem 6.36

R






E

K

F



s1 (s  1)2

Y

Sensor 1

6.37 For the system shown in Fig. 6.99, determine the Nyquist plot and apply the Nyquist criterion (a) to determine the range of values of K (positive and negative) for which the system will be stable, and (b) to determine the number of roots in the RHP for those values of K for which the system is unstable. Check your answer by using a rough root-locus sketch.

Figure 6.99 Control system for Problem 6.37

R








E

K

F

s1 (s  1)2 Sensor 1

Y

Problems

401

6.38 The Nyquist diagrams for two stable, open-loop systems are sketched in Fig. 6.100. The proposed operating gain is indicated as K0 , and arrows indicate increasing frequency. In each case give a rough estimate of the following quantities for the closed-loop (unity feedback) system: (a) (b) (c) (d)

Phase margin Damping ratio Range of gain for stability (if any) System type (0, 1, or 2)

Figure 6.100 Nyquist plots for Problem 6.38

Im[G(s)] Im[G(s)]



1 K0



1 K0 Re[G(s)]

Re[G(s)]

(a)

(b)

6.39 The steering dynamics of a ship are represented by the transfer function V (s) K[−(s/0.142) + 1] = G(s) = , δr (s) s(s/0.325 + 1)(s/0.0362) + 1) where V is the ship’s lateral velocity in meters per second, and δr is the rudder angle in radians. (a) Use the MATLAB command bode to plot the log magnitude and phase of G( jω) for K = 0.2. (b) On your plot, indicate the crossover frequency, PM, and GM. (c) Is the ship steering system stable with K = 0.2? (d) What value of K would yield a PM of 30◦ , and what would the crossover frequency be? 6.40 For the open-loop system

K(s + 1) , KG(s) = 2 s (s + 10)2

determine the value for K at the stability boundary and the values of K at the points where PM = 30◦ .

Problems for Section 6.5: Bode’s Gain–Phase Relationship 6.41 The frequency response of a plant in a unity feedback configuration is sketched in Fig. 6.101. Assume that the plant is open-loop stable and minimum-phase. (a) What is the velocity constant Kv for the system as drawn? (b) What is the damping ratio of the complex poles at ω = 100?

402 Chapter 6 The Frequency-Response Design Method Figure 6.101 Magnitude frequency response for Problem 6.41

G

100

40

10

20

1

0 db

20

0.1

0.01

1

10 20 100 v (rad/sec)

40 1000

(c) Approximately what is the system error in tracking (following) a sinusoidal input of ω = 3 rad/sec? (d) What is the PM of the system as drawn? (Estimate to within ±10◦ .) 6.42 For the system

100(s/a + 1) , s(s + 1)(s/b + 1) where b = 10a, find the approximate value of a that will yield the best PM by sketching only candidate values of the frequency-response magnitude. G(s) =

Problem for Section 6.6: Closed-Loop Frequency Response 6.43 For the open-loop system

K(s + 1) , KG(s) = 2 s (s + 10)2

determine the value for K that will yield PM ≥ 30◦ and the maximum possible closedloop bandwidth. Use MATLAB to find the bandwidth.

Problems for Section 6.7: Compensation Design 6.44 For the lead compensator D(s) =

Ts + 1 , αTs + 1

where α < 1, (a) Show that the phase of the lead compensator is given by φ = tan−1 (T ω) − tan−1 (αT ω). (b) Show that the frequency where the phase is maximum is given by ωmax =

1 √ T α

Problems

403

and that the maximum phase corresponds to sin φmax =

1−α . 1+α

(c) Rewrite your expression for ωmax to show that the maximum-phase frequency occurs at the geometric mean of the two corner frequencies on a logarithmic scale:   1 1 1 log + log . log ωmax = 2 T αT (d) To derive the same results in terms of the pole–zero locations, rewrite D(s) as D(s) =

s+z , s+p

and then show that the phase is given by     ω ω φ = tan−1 − tan−1 , |z| |p| such that ωmax =



|z||p|.

Hence the frequency at which the phase is maximum is the square root of the product of the pole and zero locations. 6.45 For the third-order servo system G(s) =

50,000 , s(s + 10)(s + 50)

use Bode plot sketches to design a lead compensator so that PM ≥ 50◦ and ωBW ≥ 20 rad/sec. Then verify and refine your design by using MATLAB. 6.46 For the system shown in Fig. 6.102, suppose that G(s) =

5 . s(s + 1)(s/5 + 1)

Use Bode plot sketches to design a lead compensation D(s) with unity DC gain so that PM ≥ 40◦ . Then verify and refine your design by using MATLAB. What is the approximate bandwidth of the system?

Figure 6.102 Control system for Problem 6.46

R






E

D

G

Y



6.47 Derive the transfer function from Td to θ for the system in Fig. 6.70. Then apply the Final Value Theorem (assuming Td = constant) to determine whether θ (∞) is nonzero for the following two cases: (a) When D(s) has no integral term: lims→0 D(s) = constant; (b) When D(s) has an integral term: D(s) = In this case, lims→0 D
(s) = constant.

D
(s) . s

404 Chapter 6 The Frequency-Response Design Method 6.48 The inverted pendulum has a transfer function given by Eq. (2.31), which is similar to 1 G(s) = 2 . s −1 (a) Use Bode plot sketches to design a lead compensator to achieve a PM of 30◦ . Then verify and refine your design by using MATLAB. (b) Sketch a root locus and correlate it with the Bode plot of the system. (c) Could you obtain the frequency response of this system experimentally? 6.49 The open-loop transfer function of a unity feedback system is G(s) =

K . s(s/5 + 1)(s/50 + 1)

(a) Use Bode plot sketches to design a lag compensator for G(s) so that the closed-loop system satisfies the following specifications: (i) The steady-state error to a unit-ramp reference input is less than 0.01. (ii) PM ≥ 40◦ . (b) Verify and refine your design by using MATLAB. 6.50 The open-loop transfer function of a unity-feedback system is G(s) =

K . s(s/5 + 1)(s/200 + 1)

(a) Use Bode plot sketches to design a lead compensator for G(s) so that the closed-loop system satisfies the following specifications: (i) The steady-state error to a unit-ramp reference input is less than 0.01. (ii) For the dominant closed-loop poles, the damping ratio ζ ≥ 0.4. (b) Verify and refine your design using MATLAB, including a direct computation of the damping of the dominant closed-loop poles. 6.51 A DC motor with negligible armature inductance is to be used in a position control system. Its open-loop transfer function is given by G(s) =

50 . s(s/5 + 1)

(a) Use Bode plot sketches to design a compensator for the motor so that the closed-loop system satisfies the following specifications: (i) The steady-state error to a unit-ramp input is less than 1/200. (ii)

The unit-step response has an overshoot of less than 20%.

(iii) The bandwidth of the compensated system is no less than that of the uncompensated system. (b) Verify and/or refine your design using MATLAB, including a direct computation of the step-response overshoot. 6.52 The open-loop transfer function of a unity-feedback system is G(s) =

K . s(1 + s/5)(1 + s/20)

(a) Sketch the system block diagram, including input reference commands and sensor noise.

Problems

405

(b) Use Bode plot sketches to design a compensator for G(s) so that the closed-loop system satisfies the following specifications: (i) The steady-state error to a unit-ramp input is less than 0.01. (ii)

PM ≥ 45◦ .

(iii) The steady-state error for sinusoidal inputs with ω < 0.2 rad/sec is less than 1/250. (iv)

Noise components introduced with the sensor signal at frequencies greater than 200 rad/sec are to be attenuated at the output by at least a factor of 100. (c) Verify and/or refine your design using MATLAB, including a computation of the closed-loop frequency response to verify (iv). 6.53

Consider a Type 1 unity-feedback system with G(s) =

K . s(s + 1)

Use Bode plot sketches to design a lead compensator so that Kv = 20 sec−1 and PM > 40◦ . Use MATLAB to verify and/or refine your design so that it meets the specifications. 6.54 Consider a satellite attitude-control system with the transfer function 0.05(s + 25) G(s) = 2 2 . s (s + 0.1s + 4) Amplitude-stabilize the system using lead compensation so that GM ≥ 2 (6 db), and PM ≥ 45◦ , keeping the bandwidth as high as possible with a single lead. 6.55 In one mode of operation, the autopilot of a jet transport is used to control altitude. For the purpose of designing the altitude portion of the autopilot loop, only the longperiod airplane dynamics are important. The linearized relationship between altitude and elevator angle for the long-period dynamics is G(s) =

20(s + 0.01) ft/ sec h(s) = . δ(s) s(s2 + 0.01s + 0.0025) deg

The autopilot receives from the altimeter an electrical signal proportional to altitude. This signal is compared with a command signal (proportional to the altitude selected by the pilot), and the difference provides an error signal. The error signal is processed through compensation, and the result is used to command the elevator actuators. A block diagram of this system is shown in Fig. 6.103. You have been given the task of designing the compensation. Begin by considering a proportional control law D(s) = K. (a) Use MATLAB to draw a Bode plot of the open-loop system for D(s) = K = 1. (b) What value of K would provide a crossover frequency (i.e., where |G| = 1) of 0.16 rad/sec? (c) For this value of K, would the system be stable if the loop were closed? (d) What is the PM for this value of K?

Figure 6.103 Control system for Problem 6.55

HR








D(s)

d

h(s) d(s)

h

1 s

H

406 Chapter 6 The Frequency-Response Design Method (e) Sketch the Nyquist plot of the system, and locate carefully any points where the phase angle is 180◦ or the magnitude is unity. (f) Use MATLAB to plot the root locus with respect to K, and locate the roots for your value of K from part (b). (g) What steady-state error would result if the command was a step change in altitude of 1000 ft? For parts (h) and (i), assume a compensator of the form D(s) = K

Ts + 1 . αTs + 1

(h) Choose the parameters K, T , and α so that the crossover frequency is 0.16 rad/sec and the PM is greater than 50◦ . Verify your design by superimposing a Bode plot of D(s)G(s)/K on top of the Bode plot you obtained for part (a), and measure the PM directly. (i) Use MATLAB to plot the root locus with respect to K for the system, including the compensator you designed in part (h). Locate the roots for your value of K from part (h). ( j) Altitude autopilots also have a mode in which the rate of climb is sensed directly and commanded by the pilot. (i)

Sketch the block diagram for this mode.

(ii)

Define the pertinent G(s).

(iii)

Design D(s) so that the system has the same crossover frequency as the altitude hold mode and the PM is greater than 50◦ .

6.56 For a system with open-loop transfer function G(s) =

10 , s[(s/1.4) + 1][(s/3) + 1]

design a lag compensator with unity DC gain so that PM ≥ 40◦ . What is the approximate bandwidth of this system? 6.57 For the ship-steering system in Problem 6.39, (a) Design a compensator that meets the following specifications: (i)

Velocity constant Kv = 2,

(ii)

PM ≥ 50◦ ,

(iii)

Unconditional stability (PM > 0 for all ω ≤ ωc , the crossover frequency).

(b) For your final design, draw a root locus with respect to K, and indicate the location of the closed-loop poles. 6.58 Consider a unity-feedback system with G(s) =

1 * +. s (s/20 + 1) s2 /1002 + 0.5s/100 + 1

(6.78)

(a) A lead compensator is introduced with α = 1/5 and a zero at 1/T = 20. How must the gain be changed to obtain crossover at ωc = 31.6 rad/sec, and what is the resulting value of Kv ? (b) With the lead compensator in place, what is the required value of K for a lag compensator that will readjust the gain to a Kv value of 100?

Problems

407

(c) Place the pole of the lag compensator at 3.16 rad/sec, and determine the zero location that will maintain the crossover frequency at ωc = 31.6 rad/sec. Plot the compensated frequency response on the same graph. (d) Determine the PM of the compensated design. 6.59 Golden Nugget Airlines had great success with their free bar near the tail of the airplane. (See Problem 5.39.) However, when they purchased a much larger airplane to handle the passenger demand, they discovered that there was some flexibility in the fuselage that caused a lot of unpleasant yawing motion at the rear of the airplane when in turbulence, which caused the revelers to spill their drinks. The approximate transfer function for the Dutch roll mode (Section 10.3.1) is r(s) 8.75(4s2 + 0.4s + 1) = , δr (s) (s/0.01 + 1)(s2 + 0.24s + 1) where r is the airplane’s yaw rate and δr is the rudder angle. In performing a finite element analysis (FEA) of the fuselage structure and adding those dynamics to the Dutch roll motion, they found that the transfer function needed additional terms which reflected the fuselage lateral bending that occurred due to excitation from the rudder and turbulence. The revised transfer function is r(s) 1 8.75(4s2 + 0.4s + 1) · = , δr (s) (s/0.01 + 1)(s2 + 0.24s + 1) (s2 /ωb2 + 2ζ s/ωb + 1) where ωb is the frequency of the bending mode (= 10 rad/sec) and ζ is the bending mode damping ratio (=0.02). Most swept-wing airplanes have a “yaw damper,” which essentially feeds back yaw rate measured by a rate gyro to the rudder with a simple proportional control law. For the new Golden Nugget airplane, the proportional feedback gain K = 1, where δr (s) = −Kr(s). (6.79) (a) Make a Bode plot of the open-loop system, determine the PM and GM for the nominal design, and plot the step response and Bode magnitude of the closedloop system. What is the frequency of the lightly damped mode that is causing the difficulty? (b) Investigate remedies to quiet down the oscillations, but maintain the same lowfrequency gain in order not to affect the quality of the Dutch roll damping provided by the yaw rate feedback. Specifically, investigate each of the following, one at a time: (i) Increasing the damping of the bending mode from ζ = 0.02 to ζ = 0.04 (would require adding energy-absorbing material in the fuselage structure). (ii)

Increasing the frequency of the bending mode from ωb = 10 rad/sec to ωb = 20 rad/sec (would require stronger and heavier structural elements).

(iii) Adding a low-pass filter in the feedback—that is, replacing K in Eq. (6.79) with KD(s), where 1 D(s) = . (6.80) s/τp + 1 Pick τp so that the objectionable features of the bending mode are reduced while maintaining the PM ≥ 60◦ . (iv)

Adding a notch filter as described in Section 5.4.3. Pick the frequency of the notch zero to be at ωb , with a damping of ζ = 0.04, and pick the denominator poles to be ( s/100 + 1)2 , keeping the DC gain of the filter = 1.

408 Chapter 6 The Frequency-Response Design Method (c) Investigate the sensitivity of the preceding two compensated designs (iii and iv) by determining the effect of a reduction in the bending mode frequency of −10%. Specifically, reexamine the two designs by tabulating the GM, PM, closed-loop bending mode damping ratio, and resonant-peak amplitude, and qualitatively describe the differences in the step response. (d) What do you recommend to Golden Nugget to help their customers quit spilling their drinks? (Telling them to get back in their seats is not an acceptable answer for this problem! Make the recommendation in terms of improvements to the yaw damper.)



6.60 Consider a system with the open-loop transfer function (loop gain) G(s) =

1 . s(s + 1)(s/10 + 1)

(a) Create the Bode plot for the system, and find GM and PM. (b) Compute the sensitivity function and plot its magnitude frequency response. (c) Compute the vector margin (VM).



6.61 Prove that the sensitivity function
(s) has magnitude greater than 1 inside a circle with a radius of 1 centered at the −1 point. What does this imply about the shape of the Nyquist plot if closed-loop control is to outperform open-loop control at all frequencies?



6.62 Consider the system in Fig. 6.102 with the plant transfer function G(s) =

10 . s(s/10 + 1)

(a) We wish to design a compensator D(s) that satisfies the following design specifications: (i) Kv = 100 (ii)

PM ≥ 45◦

(iii)

Sinusoidal inputs of up to 1 rad/sec to be reproduced with ≤ 2% error

(iv)

Sinusoidal inputs with a frequency of greater than 100 rad/sec to be attenuated at the output to ≤ 5% of their input value (b) Create the Bode plot of G(s), choosing the open-loop gain so that Kv = 100. (c) Show that a sufficient condition for meeting the specification on sinusoidal inputs is that the magnitude plot lies outside the shaded regions in Fig. 6.104. Recall that

1000

Magnitude, L(jv)

Figure 6.104 Control system constraints for Problem 6.62

W1

100 10 1 0.1 0.01

0.01

0.1

1

10 100 1000 v (rad/sec) W21

Problems

409

KG E 1 Y = and = . R 1 + KG R 1 + KG (d) Explain why introducing a lead network alone cannot meet the design specifications. (e) Explain why a lag network alone cannot meet the design specifications. (f) Develop a full design using a lead–lag compensator that meets all the design specifications without altering the previously chosen low-frequency open-loop gain.



Problems for Section 6.8: Time Delay 6.63 Assume that the system

e−Td s s + 10 has a 0.2-sec time delay (Td = 0.2 sec). While maintaining a phase margin ≥40◦ , find the maximum possible bandwidth by using the following: G(s) =

(a) One lead-compensator section D(s) = K

s+a , s+b

where b/a = 100; (b) Two lead-compensator sections  D(s) = K

 s+a 2 , s+b

where b/a = 10. (c) Comment on the statement in the text about the limitations on the bandwidth imposed by a delay. 6.64 Determine the range of K for which the following systems are stable: −4s

(a) G(s) = K e s −s (b) G(s) = K e s(s + 2) 6.65 In Chapter 5, we used various approximations for the time delay, one of which is the first order Padé: 1 − Td s/2 . e−Td s ∼ = H1 (s) = 1 + Td s/2 Using frequency response methods, the exact time delay H2 (s) = e−Td s can be obtained. Plot the phase of H1 (s) and H2 (s), and discuss the implications. 6.66 Consider the heat exchanger of Example 2.15 with the open-loop transfer function G(s) =

e−5s . (10s + 1)(60s + 1)

(a) Design a lead compensator that yields PM ≥ 45◦ and the maximum possible closedloop bandwidth. (b) Design a PI compensator that yields PM ≥ 45◦ and the maximum possible closedloop bandwidth.

410 Chapter 6 The Frequency-Response Design Method 

Problems for Section 6.9: Alternative Presentations of Data 6.67 A feedback control system is shown in Fig. 6.105. The closed-loop system is specified to have an overshoot of less than 30% to a step input. (a) Determine the corresponding PM specification in the frequency domain and the corresponding closed-loop resonant-peak value Mr . (See Fig. 6.38.) (b) From Bode plots of the system, determine the maximum value of K that satisfies the PM specification. (c) Plot the data from the Bode plots [adjusted by the K obtained in part (b)] on a copy of the Nichols chart in Fig. 6.84, and determine the resonant peak magnitude Mr . Compare that with the approximate value obtained in part (a). (d) Use the Nichols chart to determine the resonant-peak frequency ωr and the closedloop bandwidth.

Figure 6.105 Control system for Problem 6.67

R








K s(s  2)

Y

6.68 The Nichols plots of an uncompensated and a compensated system are shown in Fig. 6.106. (a) (b) (c) (d)

What are the resonance peaks of each system? What are the PM and GM of each system? What are the bandwidths of each system? What type of compensation is used?

6.69 Consider the system shown in Fig. 6.97. (a) Construct an inverse Nyquist plot of [Y ( jω)/E( jω)]−1 . (See Appendix W6.) (b) Show how the value of K for neutral stability can be read directly from the inverse Nyquist plot. (c) For K = 4, 2, and 1, determine the gain and phase margins. (d) Construct a root-locus plot for the system, and identify corresponding points in the two plots. To what damping ratios ζ do the GM and PM of part (c) correspond? 6.70 An unstable plant has the transfer function Y (s) s+1 = . F(s) (s − 1)2 A simple control loop is to be closed around it, in the same manner as in the block diagram in Fig. 6.97. (a) Construct an inverse Nyquist plot of Y /F. (See Appendix W6.)

Problems

5

0

0.9

v 25

5

0

0

1.3

v 20

0

3

0.80

1.50 v 40

2.0

2

1.0 0.9 0.8 0.7 0.6 0.5

0.70  20

v 50

5.0

Closedloop magnitude

1

1.2

4

D(s)G(s)

5 0.9



5

0

1.00

1.1

5

2 

Compensated

1.0

1

Uncompensated 10 9 8 7 6

411

0.60



0.50

3

v 70

0

0.40

v 30

0.30

0.4 0.3

0.20

0.2

0.15

Closedloop phase

v 40

0.10 0.1 180

160

140

120 100 ⬔D(s)G(s)

80

60

40

Figure 6.106 Nichols plots for Problem 6.68

(b) Choose a value of K to provide a PM of 45◦ . What is the corresponding GM? (c) What can you infer from your plot about the stability of the system when K < 0? (d) Construct a root-locus plot for the system, and identify corresponding points in the two plots. In this case, to what value of ζ does PM = 45◦ correspond?

412 Chapter 6 The Frequency-Response Design Method 6.71 Consider the system shown in Fig. 6.107(a). (a) Construct a Bode plot for the system. (b) Use your Bode plot to sketch an inverse Nyquist plot. (See Appendix W6.) (c) Consider closing a control loop around G(s), as shown in Fig. 6.107(b). Using the inverse Nyquist plot as a guide, read from your Bode plot the values of GM and PM when K = 0.7, 1.0, 1.4, and 2. What value of K yields PM = 30◦ ? (d) Construct a root-locus plot, and label the same values of K on the locus. To what value of ζ does each pair of PM/GM values correspond? Compare ζ versus PM with the rough approximation in Fig. 6.59.

Figure 6.107 Control system for Problem 6.71

U

G(s) 

4 s(s  2)2

Y

R






E

K

U

G(s)

 Sensor 1

(a)

(b)

Y

7

State-Space Design

A Perspective on State-Space Design In addition to the transform techniques of root locus and frequency response, there is a third major method of designing feedback control systems: the statespace method. We will introduce the state-variable method of describing differential equations. In state-space design, the control engineer designs a dynamic compensation by working directly with the state-variable description of the system. Like the transform techniques, the aim of the state-space method is to find a compensation D(s), such as that shown in Fig. 7.1, that satisfies the design specifications. Because the state-space method of describing the plant and computing the compensation is so different from the transform techniques, it may seem at first to be solving an entirely different problem. We selected the examples and analysis given toward the end of this chapter to help convince you that, indeed, state-space design results in a compensator with a transfer function D(s) that is equivalent to those D(s) compensators obtained with the other two methods. Because it is particularly well suited to the use of computer techniques, state-space design is increasingly studied and used today by control engineers.

Chapter Overview This chapter begins by considering the purposes and advantages of using state-space design. We discuss selection of state-variables and state-space models for various dynamic systems through several examples in Section 7.2.

413

414 Chapter 7 State-Space Design Figure 7.1 A control system design definition

R



Compensation






E

Plant U

D(s)

G(s)

Y

Models in state-variable form enhance our ability to apply the computational efficiency of computer-aided design tools such as MATLAB® . In Section 7.3 we show that it is beneficial to look at the state-variable form in terms of an analog computer simulation model. In Section 7.4 we review the development of state-variable equations from block diagrams. We then solve for the dynamic response, using state equations for both hand and computer analysis. Having covered these preliminary fundamentals, we next proceed to the major task of control system design via state-space. The steps of the design method are as follows: 1. Select closed-loop pole (root as referred to in previous chapters) locations and develop the control law for the closed-loop system that corresponds to satisfactory dynamic response (Sections 7.5 and 7.6). 2. Design an estimator (Section 7.7). 3. Combine the control law and the estimator (Section 7.8). 4. Introduce the reference input (Sections 7.5.2 and 7.9). After working through the central design steps, we briefly explore the use of integral control in state-space (Section 7.10). The next three sections of this chapter consider briefly some additional concepts pertaining to the state-space method; because they are relatively advanced, they may be considered optional to some courses or readers. Finally Section 7.14 provides some historical perspective for the material in this chapter.

7.1 Advantages of State-Space

Normal form

The idea of state-space comes from the state-variable method of describing differential equations. In this method the differential equations describing a dynamic system are organized as a set of first-order differential equations in the vector-valued state of the system, and the solution is visualized as a trajectory of this state vector in space. State-space control design is the technique in which the control engineer designs a dynamic compensation by working directly with the state-variable description of the system. Thus far, we have seen that the ordinary differential equations (ODEs) of physical dynamic systems can be manipulated into state-variable form. In the field of mathematics, where ODEs are studied, the state-variable form is called the normal form for the equations. There are several good reasons for studying equations in this form, three of which are listed here: •

To study more general models: The ODEs do not have to be linear or stationary. Thus, by studying the equations themselves, we can develop methods that are very general. Having them in state-variable form gives us a compact, standard

7.1 Advantages of State-Space

• Phase plane

•

415

form for study. Furthermore, the techniques of state-space analysis and design easily extend to systems with multiple inputs and/or multiple outputs. Of course, in this text we study mainly linear time-invariant models with single input and output (for the reasons given earlier). To introduce the ideas of geometry into differential equations: In physics the plane of position versus velocity of a particle or rigid body is called the phase plane, and the trajectory of the motion can be plotted as a curve in this plane. The state is a generalization of that idea to include more than two dimensions. While we cannot plot more than three dimensions, the concepts of distance, of orthogonal and parallel lines, and other concepts from geometry can be useful in visualizing the solution of an ODE as a path in state-space. To connect internal and external descriptions: The state of a dynamic system often directly describes the distribution of internal energy in the system. For example, it is common to select the following as state-variables: position (potential energy), velocity (kinetic energy), capacitor voltage (electric energy), and inductor current (magnetic energy). The internal energy can always be computed from the state-variables. By a system of analysis to be described shortly, we can relate the state to the system inputs and outputs and thus connect the internal variables to the external inputs and to the sensed outputs. In contrast, the transfer function relates only the input to the output and does not show the internal behavior. The state form keeps the latter information, which is sometimes important.

Use of the state-space approach has often been referred to as modern control design, and use of transfer-function-based methods, such as root locus and frequency response, referred to as classical control design. However, because the state-space method of description for ODEs has been in use for over 100 years and was introduced to control design in the late 1950s, it seems somewhat misleading to refer to it as modern. We prefer to refer to the two approaches to design as state-space methods and transform methods. Advantages of state-space design are especially apparent when the system to be controlled has more than one control input or more than one sensed output. However, in this book we shall examine the ideas of state-space design using the simpler singleinput-single output (SISO) systems. The design approach used for systems described in state form is “divide and conquer.” First, we design the control as if all of the state were measured and available for use in the control law. This provides the possibility of assigning arbitrary dynamics for the system. Having a satisfactory control law based on full-state feedback, we introduce the concept of an observer and construct estimates of the state based on the sensed output. We then show that these estimates can be used in place of the actual state-variables. Finally, we introduce the external reference-command inputs, and the structure is complete. Only at this point can we recognize that the resulting compensation has the same essential structure as that developed with transform methods. Before we can begin the design using state descriptions, it is necessary to develop some analytical results and tools from matrix linear algebra for use throughout the chapter. We assume that you are familiar with such elementary matrix concepts as the identity matrix, triangular and diagonal matrices, and the transpose of a matrix.

416 Chapter 7 State-Space Design We also assume that you have some familiarity with the mechanics of matrix algebra, including adding, multiplying, and inverting matrices. More advanced results will be developed in Section 7.4 in the context of the dynamic response of a linear system. All of the linear algebra results used in this chapter are repeated in Appendix WE for your reference and review.

7.2 System Description in State-Space The motion of any finite dynamic system can be expressed as a set of first-order ODEs. This is often referred to as the state-variable representation. For example, the use of Newton’s law and the free-body diagram in Section 2.1 typically lead to second-order differential equations—that is, equations that contain the second derivative, such as x¨ in Eq. (2.3) or θ¨ in Eq. (2.15). The latter equation can be expressed as x˙ 1 = x2 , u x˙ 2 = , I

(7.1) (7.2)

where u = Fc d + MD , x1 = θ , x2 = θ˙ , ¨ x˙ 2 = θ. Standard form of linear differential equations

The output of this system is θ, the satellite attitude. These same equations can be represented in the state-variable form as the vector equation x˙ = Fx + Gu, (7.3) where the input is u and the output is y = Hx + Ju.

(7.4)

The column vector x is called the state of the system and contains n elements for an nth-order system. For mechanical systems, the state vector elements usually consist of the positions and velocities of the separate bodies, as is the case for the example in Eqs. (7.1) and (7.2). The quantity F is an n × n system matrix, G is an n × 1 input matrix, H is a 1 × n row matrix referred to as the output matrix, and J is a scalar called the direct transmission term.1 To save space, we will sometimes refer to a state vector by its transpose, x = [ x1 which is equivalent to

x2 . . . ]T ,

⎡

⎤ x1 ⎢ ⎥ x = ⎣ x2 ⎦ . .. .

1 It is also common to use the notation A, B, C, and D in place of F, G, H, and J. We will typically use F,

G to represent plant dynamics and A, B to represent a general linear system.

7.2 System Description in State-Space

417

The differential equation models of more complex systems, such as those developed in Chapter 2 on mechanical, electrical, and electromecheanical systems, can be described by state-variables through selection of positions, velocities, capacitor voltages, and inductor currents as suitable state-variables. In this chapter we will consider control systems design using the state-variable form. For the case in which the relationships are nonlinear [such as the case in Eqs. (2.22), (2.75), and (2.79)], the linear form cannot be used directly. One must linearize the equations as we did in Chapter 2 to fit the form (see also Chapter 9). The state-variable method of specifying differential equations is used by computer-aided control systems design software packages (e.g., MATLAB). Therefore, in order to specify linear differential equations to the computer, you need to know the values of the matrices F, G, H, and the constant J.

EXAMPLE 7.1

Satellite Attitude Control Model in State-Variable Form Determine the F, G, H, J matrices in the state-variable form for the satellite attitude control model in Example 2.3 with MD = 0. Solution. Define the attitude and the angular velocity of the satellite as the statevariables so that x  [θ ω]T.2 The single second-order equation (2.15) can then be written in an equivalent way as two first-order equations: θ˙ = ω, d ω˙ = Fc . I These equations are expressed, using Eq. (7.3), x˙ = Fx + Gu, as        0 1 θ 0 θ˙ = + Fc . 0 0 ω d/I ω˙ The output of the system is the satellite attitude, y = θ. Using Eq. (7.4), y = Hx + Ju, this relation is expressed as   θ y=[ 1 0 ] . ω Therefore, the matrices for the state-variable form are     0 1 0 F= , G= , H=[ 1 0 0 d/I

0 ],

J = 0,

and the input u  Fc . For this very simple example, the state-variable form is a more cumbersome way of writing the differential equation than the second-order version in Eq. (2.15). However, the method is not more cumbersome for most systems, and the advantages of having a standard form for use in computer-aided design have led to widespread use of the state-variable form.

2 The symbol

 means “is to be defined.”

418 Chapter 7 State-Space Design The next example has more complexity and shows how to use MATLAB to find the solution of linear differential equations.

EXAMPLE 7.2

Cruise Control Step Response (a) Rewrite the equation of motion from Example 2.1 in state-variable form, where the output is the car position x. (b) Use MATLAB to find the response of the velocity of the car for the case in which the input jumps from being u = 0 at time t = 0 to a constant u = 500 N thereafter. Assume that the car mass m is 1000 kg and b = 50 N·sec/m. Solution (a) Equations of motion: First we need to express the differential equation describing the plant, Eq. (2.3), as a set of simultaneous first-order equations. To do so, we define the position and the velocity of the car as the state-variables x and v, so that x = [x v]T . The single second-order equation, Eq. (2.3), can then be rewritten as a set of two first-order equations: x˙ = v, b 1 v˙ = − v + u. m m Next, we use the standard form of Eq. (7.3), x˙ = Fx + Gu, to express these equations:        x˙ 0 1 x 0 = + u. (7.5) v˙ 0 −b/m v 1/m The output of the system is the car position y = x1 = x, which is expressed in matrix form as   x y=[ 1 0 ] , v or y = Hx. So the state-variable-form matrices defining this example are     0 1 0 F= , G= , H = [ 1 0 ], J = 0. 0 −b/m 1/m (b) Time response: The equations of motion are those given in part (a), except that now the output is v = x2 . Therefore, the output matrix is H=[ 0

1 ].

The coefficients required are b/m = 0.05 and 1/m = 0.001. The numerical values of the matrices defining the system are thus     0 1 0 F= , G= , H = [ 0 1 ], J = 0. 0 −0.05 0.001 The step function in MATLAB calculates the time response of a linear system to a unit-step input. Because the system is linear, the output for this case can be multiplied by the magnitude of the input step to derive a step response of any amplitude. Equivalently, the G matrix can be multiplied by the magnitude of the input step.

7.2 System Description in State-Space Figure 7.2 Response of the car velocity to a step in u

419

10

Amplitude

8 6 4 2 0

Step response with MATLAB

0

50 Time (sec)

100

The statements F = [0 1;0 −0.05]; G = [0;0.001]; H = [0 1]; J = 0; sys = ss(F, 500*G,H,J); % step gives unit step response, so 500*G gives u = 500 N. step(sys); % plots the step response calculate and plot the time response for an input step with a 500-N magnitude. The step response is shown in Fig. 7.2.

EXAMPLE 7.3

Bridged Tee Circuit in State-Variable Form Determine the state-space equations for the circuit shown in Fig. 2.25. Solution. In order to write the equations in the state-variable form (i.e., a set of simultaneous first-order differential equations), we select the capacitor voltages v1 and v2 as the state elements (i.e., x = [v1 v2 ]T ) and vi as the input (i.e., u = vi ). Here v1 = v2 , v2 = v1 − v3 , and still v1 = vi . Thus v1 = vi , v2 = v1 , and v3 = vi − v2 . In terms of v1 and v2 , Eq. (2.34) is dv v − (vi − v2 ) v 1 − vi + 1 + C1 1 = 0. R1 R2 dt Rearranging this equation into standard form, we get       dv1 1 1 1 1 1 1 1 1 =− + + v1 − v2 + vi . dt C1 R1 R2 C1 R2 C1 R1 R2 In terms of v1 and v2 , Eq. (2.35) is vi − v2 − v1 d + C2 (vi − v2 − vi ) = 0. R2 dt

(7.6)

420 Chapter 7 State-Space Design In standard form, the equation is dv2 v vi v =− 1 − 2 + . (7.7) dt C2 R2 C2 R2 C2 R2 Equations (2.34)–(2.35) are entirely equivalent to the state-variable form, Eqs. (7.6) and (7.7), in describing the circuit. The standard matrix definitions are

⎤ ⎡ − C1 R1 + R1 − C1 R1 2 1 2 ⎦ F=⎣ 1 1 , − C 1R − C 1R ⎡

1

G = ⎣ C1  H= 0

EXAMPLE 7.4




2 2

1 + R1 R1 2 1 C2 R2

⎤

2 2

⎦,

−1 , J = 1.

Loudspeaker with Circuit in State-Variable Form For the loudspeaker in Fig. 2.29 and the circuit driving it in Fig. 2.30 find the statespace equations relating the input voltage va to the output cone displacement x. Assume that the effective circuit resistance is R and the inductance is L. Solution. Recall the two coupled equations, (2.44) and (2.48), that constitute the dynamic model for the loudspeaker: M x¨ + b˙x = 0.63i, di L + Ri = va − 0.63˙x . dt A logical state vector for this third-order system would be x  [x x˙ i]T , which leads to the standard matrices ⎡ ⎤ ⎡ ⎤ 0 1 0 0  0.63/M ⎦ , G = ⎣ 0 ⎦ , H = 1 0 0 , J = 0, F = ⎣ 0 −b/M 0 −0.63/L −R/L 1/L where now the input u  va .

EXAMPLE 7.5

Modeling a DC Motor in State-Variable Form Find the state-space equations for the DC motor with the equivalent electric circuit shown in Fig. 2.32(a). Solution. Recall the equations of motion [Eqs. (2.52) and (2.53)] from Chapter 2: Jm θ¨m + bθ˙m = Kt ia , dia + Ra ia = va − Ke θ˙m . La dt

7.3 Block Diagrams and State-Space 421 T  A state vector for this third-order system is x  θm θ˙m ia , which leads to the standard matrices ⎡ ⎤ ⎡ ⎤ 0 0 1 0 Kt ⎥ ⎢ ⎥ ⎢ 0 −b  ⎥ G = ⎢ 0 ⎥ , H = 1 0 0 , J = 0, F=⎢ J J m m ⎣ ⎦ ⎣ ⎦ 1 e a 0 −K −R La L L a

a

where the input u  va . The state-variable form can be applied to a system of any order. Example 7.6 illustrates the method for a fourth-order system.

EXAMPLE 7.6

Flexible Disk Drive in State-Variable Form Find the state-variable form of the differential equations for Example 2.4, where the output is θ2 . Solution. Define the state vector to be  T x = θ1 θ˙1 θ2 θ˙2 . Then solve Eqs. (2.17) and (2.18) for θ¨1 and θ¨2 so that the state-variable form is more apparent. The resulting matrices are ⎡ ⎤ ⎡ ⎤ 0 1 0 0 0 k b ⎥ ⎢ 1 ⎥ ⎢ −k −b ⎢ ⎥ ⎢ I1 ⎥ I I I 1 1 1 F=⎢ ⎥ , G = ⎢ I1 ⎥ , 0 0 1 ⎦ ⎣ 0 ⎦ ⎣ 0 k b 0 − Ik − Ib I2 I2 2 2  H = 0 0 1 0 , J = 0. Difficulty arises if the differential equation contains derivatives of the input u. Techniques to handle this situation will be discussed in Section 7.4.

7.3 Block Diagrams and State-Space Perhaps the most effective way of understanding the state-variable equations is via an analog computer block-diagram representation. The structure of the representation uses integrators as the central element, which are quite suitable for first-order, state-variable representation of equations of motion for a system. Even though the analog computers are almost extinct, analog computer implementation is still a useful concept for state-variable design, and in the circuit design of analog compensation.3 The analog computer was a device composed of electric components designed to simulate ODEs. The basic dynamic component of the analog computer is an integrator, constructed from an operational amplifier with a capacitor feedback and a resistor feed-forward as shown in Fig. 2.28. Because an integrator is a device whose 3As well as due to its historical significance.

422 Chapter 7 State-Space Design Figure 7.3 An integrator Figure 7.4 Components of an analog computer

x

1 s

x

IC

Integrator

Summer

e1 e2

10

e0  (10e1  e2) dt  IC t

0

1

e1

10

e2

1

Potentiometer e1

e0  (10e1  e2)

k

e0  ke1 0  k  1

input is the derivative of its output, as shown in Fig. 7.3, if, in an analog-computer simulation, we identify the outputs of the integrators as the state, we will then automatically have the equations in state-variable form. Conversely, if a system is described by state-variables, we can construct an analog-computer simulation of that system by taking one integrator for each state-variable and connecting its input according to the given equation for that state-variable as expressed in the state-variable equations. The analog-computer diagram is a picture of the state equations. The components of a typical analog computer used to accomplish these functions are shown in Fig. 7.4. Notice that the operational amplifier has a sign change that gives it a negative gain.

EXAMPLE 7.7

Analog-Computer Implementation Find a state-variable description and the transfer function of the third-order system shown in Fig. 7.5 whose differential equation is ... y + 6¨y + 11˙y + 6y = 6u. Solution. We solve for the highest derivative term in the ODE to obtain ... y = −6¨y − 11˙y − 6y + 6u.

(7.8)

Now we assume that we have this highest derivative and note that the lower order terms can be obtained by integration as shown in Fig. 7.6(a). Finally, we apply Eq. (7.8) to complete the realization shown in Fig. 7.6(b). To obtain the state description, we simply define the state-variables as the output of the integrators x1 = y¨ , x2 = y˙ , x3 = y, to obtain x˙ 1 = −6x1 − 11x2 − 6x3 + 6u, x˙ 2 = x1 , x˙ 3 = x2 ,

7.3 Block Diagrams and State-Space Figure 7.5 Third-order system

u

6



1 s




  

x1

1 s

x2

x3

1 s

423

y

6 11 6

Figure 7.6 Block diagram of a system to solve ... y +6¨y +11˙y +6y = 6u, using only integrators as dynamic elements: (a) intermediate diagram; (b) final diagram

y

y

1 s

y

1 s

1 s

y

(a)

u

6

 




y  

y

1 s

1 s

y

1 s

y

6 11 6 (b)

which provides the state-variable description ⎡ ⎤ ⎡ ⎤ −6 −11 −6 6 0 0 ⎦, G = ⎣ 0 ⎦, F=⎣ 1 0 1 0 0

H=[ 0

0

1 ],

J = 0.

The MATLAB statement [num,den] = ss2tf(F,G,H,J); will yield the transfer function 6 Y (s) = 3 . U(s) s + 6s2 + 11s + 6 If the transfer function were desired in factored form, it could be obtained by transforming either the ss or tf description. Therefore, either of the MATLAB statements % convert state-variable realization to pole–zero form [z,p,k] = ss2zp(F,G,H,J) and % convert numerator-denominator to pole–zero form [z,p,k] = tf2zp(num,den) would result in z = [], p =



−3

−2

−1




, k = 6.

424 Chapter 7 State-Space Design This means that the transfer function could also be written in factored form as Y (s) 6 = . U(s) (s + 1)(s + 2)(s + 3)

7.3.1 Time and Amplitude Scaling in State-Space We have already discussed time and amplitude scaling in Chapter 3 . We now extend the ideas to the state-variable form. Time scaling with τ = ωo t replaces Eq. (7.3) with dx 1 1 = Fx + Gu = F
x + G
u. (7.9) dτ ωo ωo Amplitude scaling of the state corresponds to replacing x with z = D−1 x x, where Dx is a diagonal matrix of scale factors. Input scaling corresponds to replacing u with v = D−1 u u. With these substitutions, Dx z˙ =

1 1 FDx z + GDu v. ωo ωo

(7.10)

Then z˙ =

1 −1 1 −1 D FDx z + D GDu v = F
z + G
v. ωo x ωo x

(7.11)

Equation (7.11) compactly expresses the time- and amplitude-scaling operations. Regrettably, it does not relieve the engineer of the responsibility of actually thinking of good scale factors so that scaled equations are in good shape.

EXAMPLE 7.8

Time Scaling an Oscillator The equation for an oscillator was derived in Example 2.5. For a case with a very fast natural frequency ωn = 15,000 rad/sec (about 2 kHz), Eq. (2.23) can be rewritten as θ¨ + 150002 · θ = 106 · Tc . Determine the time-scaled equation so that the unit of time is milliseconds. ˙ T , the unscaled Solution. In state-variable form with a state vector x = [θ θ] matrices are     0 0 1 . and G = F= 106 −150002 0 Applying Eq. (7.9) results in ⎡ 0 F
= ⎣ 2 15000 − 1000

⎤ 1 1000 ⎦ 0

⎡ and G
= ⎣

which yields state-variable equations that are scaled.

⎤ 0 103

⎦,

7.4 Analysis of the State Equations

425

7.4 Analysis of the State Equations In the previous section we introduced and illustrated the process of selecting a state and organizing the equations in state form. In this section we review that process and describe how to analyze the dynamic response using the state description. In Section 7.4.1 we begin by relating the state description to block diagrams and the Laplace transform description and to consider the fact that for a given system the choice of state is not unique. We show how to use this nonuniqueness to select among several canonical forms for the one that will help solve the particular problem at hand; a control canonical form makes feedback gains of the state easy to design. After studying the structure of state equations in Section 7.4.2, we consider the dynamic response and show how transfer-function poles and zeros are related to the matrices of the state descriptions. To illustrate the results with hand calculations, we offer a simple example that represents the model of a thermal system. For more realistic examples, a computer-aided control systems design software package such as MATLAB is especially helpful; relevant MATLAB commands will be described from time to time.

7.4.1 Block Diagrams and Canonical Forms We begin with a thermal system that has a simple transfer function G=

b(s) s+2 2 −1 = 2 = + . a(s) s + 7s + 12 s+4 s+3

(7.12)

The roots of the numerator polynomial b(s) are the zeros of the transfer function, and the roots of the denominator polynomial a(s) are the poles. Notice that we have represented the transfer function in two forms, as a ratio of polynomials and as the result of a partial-fraction expansion. In order to develop a state description of this system (and this is a generally useful technique), we construct a block diagram that corresponds to the transfer function (and the differential equations) using only isolated integrators as the dynamic elements. One such block diagram, structured in control canonical form, is drawn in Fig. 7.7. The central feature of this structure is that each state-variable is connected by the feedback to the control input. Once we have drawn the block diagram in this form, we can identify the state description matrices simply by inspection; this is possible because when the output of an integrator is a state-variable, the input of that integrator is the derivative of that Figure 7.7 A block diagram representing Eq. (7.12) in control form

1

U






x1








1 s 7

 12

x 1 x2

1 s

x2

2








Y

426 Chapter 7 State-Space Design variable. For example, in Fig. 7.7, the equation for the first state-variable is x˙ 1 = −7x1 − 12x2 + u. Continuing in this fashion, we get x˙ 2 = x1 , y = x1 + 2x2 . These three equations can then be rewritten in the matrix form

where

 Ac =

(7.13)

y = Cc x,

(7.14)

−7 −12 1 0

Cc = [ 1

MATLAB tf2ss

x˙ = Ac x + Bc u,

2 ],



 ,

Bc =

1 0

 ,

Dc = 0,

(7.15a) (7.15b)

and where the subscript c refers to control canonical form. Two significant facts about this form are that the coefficients 1 and 2 of the numerator polynomial b(s) appear in the Cc matrix, and (except for the leading term) the coefficients 7 and 12 of the denominator polynomial a(s) appear (with opposite signs) as the first row of the Ac matrix. Armed with this knowledge, we can thus write down by inspection the state matrices in control canonical form for any system whose transfer function is known as a ratio of numerator and denominator polynomials. If b(s) = b1 sn−1 + b2 sn−2 + · · · + bn and a(s) = sn + a1 sn−1 + a2 sn−2 + · · · + an , then the MATLAB steps are num = b = [b1 b2 · · · bn ] den = a = [1 a1 a2 · · · an ] [Ac , Bc , Cc , Dc ] = tf2ss(num,den).

Control canonical form

We read tf2ss as “transfer function to state-space.” The result will be ⎡ ⎡ ⎤ ⎤ −a1 −a2 · · · · · · −an 1 ⎢ 1 ⎢0⎥ ⎥ 0 · · · · · · 0 ⎢ ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ 1 0 ··· 0 ⎥ Ac = ⎢ 0 Bc = ⎢0⎥ , ⎥, ⎢ .. ⎢ .. ⎥ ⎥ . . .. 0 .. ⎦ ⎣ . ⎣.⎦ 0 0··· ··· 1 0 0  Dc = 0. Cc = b1 b2 · · · · · · bn ,

(7.16a)

(7.16b)

The block diagram of Fig. 7.7 and the corresponding matrices of Eq. (7.15) are not the only way to represent the transfer function G(s). A block diagram corresponding to the partial-fraction expansion of G(s) is given in Fig. 7.8. Using the same technique as before, with the state-variables marked as shown in the figure, we can determine the matrices directly from the block diagram as being z˙ = Am z + Bm u, y = Cm z + Dm u,

7.4 Analysis of the State Equations Figure 7.8 Block diagram for Eq. (7.12) in modal canonical form



U

z1




z1

1 s





2




427

Y



4 

z2




z2

1 s



1

3

where



 0 , −3

−4 Am = 0

−1],

Cm = [2 Modal form

(7.17a)

Dm = 0,

(7.17b)

and the subscript m refers to modal canonical form. The name for this form derives from the fact that the poles of the system transfer function are sometimes called the normal modes of the system. The important fact about the matrices in this form is that the system poles (here −4 and −3) appear as the elements along the diagonal of the Am matrix, and the residues, the numerator terms in the partial-fraction expansion (here 2 and −1), appear in the Cm matrix. Expressing a system in modal canonical form can be complicated by two factors: (1) the elements of the matrices will be complex when the poles of the system are complex, and (2) the system matrix cannot be diagonal when the partial-fraction expansion has repeated poles. To solve the first problem, we express the complex poles of the partial-fraction expansion as conjugate pairs in second-order terms so that all the elements remain real. The corresponding Am matrix will then have 2×2 blocks along the main diagonal representing the local coupling between the variables of the complex-pole set. To handle the second difficulty, we also couple the corresponding state-variables, so that the poles appear along the diagonal with off-diagonal terms indicating the coupling. A simple example of this latter case is the satellite system from Example 7.1, whose transfer function is G(s) = 1/s2 . The system matrices for this transfer function in a modal form are  F=

EXAMPLE 7.9

  1 Bm = , 1

0 0

1 0



 ,

G=

0 1

 ,

H=[ 1

0 ],

J = 0.

(7.18)

State Equations in Modal Canonical Form A “quarter car model” [see Eq. (2.12)] with one resonant mode has a transfer function given by 1 1 2s + 4 = 2− 2 . (7.19) G(s) = 2 2 s (s + 2s + 4) s s + 2s + 4 Find state matrices in modal form describing this system.

428 Chapter 7 State-Space Design Figure 7.9 Block diagram for a fourth-order system in modal canonical form with shading indicating portion in control canonical form

U






x1

1 s

x1

1 s

x3

1 s

x3

1 s








x2

x4

1








Y

2

 4

Solution. The transfer function has been given in real partial-fraction form. To get state-description matrices, we draw a corresponding block diagram with integrators only, assign the state, and write down the corresponding matrices. This process is not unique, so there are several acceptable solutions to the problem as stated, but they will differ in only trivial ways. A block diagram with a satisfactory assignment of variables is given in Fig. 7.9. Notice that the second-order term to represent the complex poles has been realized in control canonical form. There are a number of other possibilities that can be used as alternatives for this part. This particular form allows us to write down the system matrices by inspection: ⎡ ⎤ ⎡ ⎤ 0 0 0 0 1 ⎢ 1 0 ⎢ 0 ⎥ 0 0 ⎥ ⎥ ⎢ ⎥ F=⎢ ⎣ 0 0 −2 −4 ⎦ , G = ⎣ 1 ⎦ , 0 0 1 0 0 H=[ 0

State description and output equation

1

0

−1 ],

J = 0.

(7.20)

Thus far, we have seen that we can obtain the state description from a transfer function in either control or modal form. Because these matrices represent the same dynamic system, we might ask as to, what is the relationship between the matrices in the two forms (and their corresponding state-variables)? More generally, suppose we have a set of state equations that describe some physical system in no particular form, and we are given a problem for which the control canonical form would be helpful. (We will see such a problem in Section 7.5.) Is it possible to calculate the desired canonical form without obtaining the transfer function first? To answer these questions requires a look at the topic of state transformations. Consider a system described by the state equations x˙ = Fx + Gu,

(7.21a)

y = Hx + Ju.

(7.21b)

As we have seen, this is not a unique description of the dynamic system. We consider a change of state from x to a new state z that is a linear transformation of x. For a nonsingular matrix T, we let x = Tz. (7.22)

7.4 Analysis of the State Equations

429

By substituting Eq. (7.22) into Eq. (7.21a), we have the equations of motion in terms of the new state z: x˙ = T˙z = FTz + Gu, (7.23a)

Transformation of state

z˙ = T−1 FTz + T−1 Gu,

(7.23b)

z˙ = Az + Bu.

(7.23c)

In Eq. (7.23c), A = T−1 FT,

(7.24a)

−1

(7.24b) B = T G. Then we substitute Eq. (7.22) into Eq. (7.21b) to get the output in terms of the new state z: y = HTz + Ju = Cz + Du. Here C = HT, D = J. (7.25) Given the general matrices F, G, and H and scalar J, we would like to find the transformation matrix T such that A, B, C, and D are in a particular form, for example, control canonical form. To find such a T, we assume that A, B, C, and D are already in the required form, further assume that the transformation T has a general form, and match terms. Here we will work out the third-order case; how to extend the analysis to the general case should be clear from the development. It goes like this. First we rewrite Eq. (7.24a) as AT−1 = T−1 F. If A is in control canonical form, and we describe T−1 as a matrix with rows t1 , t2 , and t3 , then ⎡ ⎤ ⎡ ⎤ ⎤⎡ −a1 −a2 −a3 t1 F t1 ⎣ 1 0 0 ⎦ ⎣ t2 ⎦ = ⎣ t2 F ⎦ . (7.26) t3 0 1 0 t3 F Working out the third and second rows gives the matrix equations t2 = t3 F,

(7.27a)

(7.27b) t1 = t2 F = t3 F . From Eq. (7.24b), assuming that B is also in control canonical form, we have the relation T−1 G = B, or ⎡ ⎤ ⎡ ⎤ 1 t1 G ⎣ t2 G ⎦ = ⎣ 0 ⎦ . (7.28) t3 G 0 2

Combining Eqs. (7.27) and (7.28), we get t3 G = 0, t2 G = t3 FG = 0, t1 G = t3 F2 G = 1.

430 Chapter 7 State-Space Design These equations can in turn be written in matrix form as t3 [ G

FG

F2 G ] = [ 0

or t3 = [ 0 Controllability matrix transformation to control canonical form

0

0

1 ]

1 ]C −1 ,

(7.29)

]. Having t3 , we can now where the controllability matrix C = [ G FG go back to Eq. (7.27) and construct all the rows of T−1 . To sum up, the recipe for converting a general state description of dimension n to control canonical form is as follows: F2 G

•

From F and G, form the controllability matrix C=[ G

•

FG

(7.30)

Compute the last row of the inverse of the transformation matrix as tn = [ 0

•

· · · Fn−1 G ].

0

· · · 1 ]C −1 .

Construct the entire transformation matrix as ⎡ tn Fn−1 ⎢ tn Fn−2 ⎢ T−1 = ⎢ .. ⎣ .

(7.31)

⎤ ⎥ ⎥ ⎥. ⎦

(7.32)

tn • Controllable systems

Compute the new matrices from T−1 , using Eqs. (7.24a), (7.24b), and (7.25).

When the controllability matrix C is nonsingular, the corresponding F and G matrices are said to be controllable. This is a technical property that usually holds for physical systems and will be important when we consider feedback of the state in Section 7.5. We will also consider a few physical illustrations of loss of controllability at that time. Because computing the transformation given by Eq. (7.32) is numerically difficult to do accurately, it is almost never done. The reason for developing this transformation in some detail is to show how such changes of state could be done in theory and to make the following important observation: One can always transform a given state description to control canonical form if (and only if) the controllability matrix C is nonsingular. If we need to test for controllability in a real case with numbers, we use a numerically stable method that depends on converting the system matrices to “staircase” form rather than on trying to compute the controllability matrix. Problem 7.29 at the end of the chapter calls for consideration of this method. An important question regarding controllability follows directly from our discussion so far: What is the effect of a state transformation on controllability? We can show the result by using Eqs. (7.30), (7.24a), and (7.24b). The controllability matrix of the system (F, G) is Cx = [ G

FG

· · · Fn−1 G ].

(7.33)

7.4 Analysis of the State Equations

431

After the state transformation, the new description matrices are given by Eqs. (7.24a) and (7.24b), and the controllability matrix changes to AB · · · An−1 B ]

Cz = [ B =[

T−1 G

T−1 FTT−1 G

(7.34a) ···

T−1 Fn−1 TT−1 G

= T−1 Cx .

]

(7.34b) (7.34c)

Thus we see that Cz is nonsingular if and only if Cx is nonsingular, yielding the following observation: A change of state by a nonsingular linear transformation does not change controllability.

Observer canonical form

We return once again to the transfer function of Eq. (7.12), this time to represent it with the block diagram having the structure known as observer canonical form (Fig. 7.10). The corresponding matrices for this form are     −7 1 1 , Bo = , (7.35a) Ao = −12 0 2 Co = [ 1

0 ],

Do = 0.

(7.35b)

The significant fact about this canonical form is that all the feedback is from the output to the state-variables. Let us now consider what happens to the controllability of this system as the zero at −2 is varied. For this purpose, we replace the second element 2 of Bo with the variable zero location −zo and form the controllability matrix: Cx = [ Bo Ao Bo ]   1 −7 − zo . = −zo −12 The determinant of this matrix is a function of zo :

(7.36a) (7.36b)

det(Cx ) = −12 + (zo )(−7 − zo ) = −(zo2 + 7zo + 12). This polynomial is zero for zo = −3 or −4, implying that controllability is lost for these values. What does this mean? In terms of the parameter zo , the transfer function is Figure 7.10 Observer canonical form

U 2 


 12

x2

1 s

x2 




 7

x1

1 s

x1

Y

432 Chapter 7 State-Space Design s − zo . (s + 3)(s + 4) If zo = −3 or −4, there is a pole–zero cancellation and the transfer function reduces from a second-order system to a first-order one. When zo = −3, for example, the mode at −3 is decoupled from the input and control of this mode is lost. Notice that we have taken the transfer function given by Eq. (7.12) and given it two realizations, one in control canonical form and one in observer canonical form. The control form is always controllable for any value of the zero, while the observer form loses controllability if the zero cancels either of the poles. Thus, these two forms may represent the same transfer function, but it may not be possible to transform the state of one to the state of the other (in this case, from observer to control canonical form). While a transformation of state cannot affect controllability, the particular state selected from a transfer function can: G(s) =

Controllability is a function of the state of the system and cannot be decided from a transfer function. To discuss controllability more at this point would take us too far afield. The closely related property of observability and the observer canonical form will be taken up in Section 7.7.1. A more detailed discussion of these properties of dynamic systems is given in the Appendix WF, for those who would like to learn more. We return now to the modal form for the equations, given by Eqs. (7.17a) and (7.17b) for the example transfer function. As mentioned before, it is not always possible to find a modal form for transfer functions that have repeated poles, so we assume our system has only distinct poles. Furthermore, we assume that the general state equations given by Eqs. (7.21a) and (7.21b) apply. We want to find a transformation matrix T defined by Eq. (7.22) such that the transformed Eqs. (7.24a) and (7.25) will be in modal form. In this case, we assume that the A matrix is diagonal and that T is composed of the columns t1 , t2 , and t3 . With this assumption, the state transformation Eq. (7.24a) becomes



Transformation to modal form

Eigenvectors Eigenvalues

⎡

t1

t2

p1 t3 ⎣ 0 0

TA = FT ⎤ 0 0 p2 0 ⎦ = F[ t1 0 p3

t2

t3 ].

(7.37)

Equation (7.37) is equivalent to the three vector-matrix equations pi ti = Fti ,

i = 1, 2, 3.

(7.38)

In matrix algebra Eq. (7.38) is a famous equation, whose solution is known as the eigenvector/eigenvalue problem. Recall that ti is a vector, F is a matrix, and pi is a scalar. The vector ti is called an eigenvector of F, and pi is called the corresponding eigenvalue. Because we saw earlier that the modal form is equivalent to a partialfraction-expansion representation with the system poles along the diagonal of the state matrix, it should be clear that these “eigenvalues” are precisely the poles of our system. The transformation matrix that will convert the state description matrices to modal form has as its columns the eigenvectors of F, as shown in Eq. (7.37) for

7.4 Analysis of the State Equations

MATLAB eig

EXAMPLE 7.10

433

the third-order case. As it happens, there are robust, reliable computer algorithms to compute eigenvalues and the eigenvectors of quite large systems using the QR algorithm.4 In MATLAB, the command p = eig(F) is the way to compute the poles if the system equations are in state form. Notice also that Eq. (7.38) is homogeneous in that, if ti is an eigenvector, so is αti , for any scalar α. In most cases the scale factor is selected so that the length (square root of the sum of squares of the magnitudes of the elements) is unity. MATLAB will perform this operation. Another option is to select the scale factors so that the input matrix B is composed of all 1’s. The latter choice is suggested by a partial-fraction expansion with each part realized in control canonical form. If the system is real, then each element of F is real, and if p = σ + jω is a pole, so is the conjugate, p∗ = σ − jω. For these eigenvalues, the eigenvectors are also complex and conjugate. It is possible to compose the transformation matrix using the real and complex parts of the eigenvectors separately, so the modal form is real but has 2 × 2 blocks for each pair of complex poles. Later, we will see the result of the MATLAB function that does this, but first let us look at the simple real-poles case.

Transformation of Thermal System from Control to Modal Form Find the matrix to transform the control form matrices in Eq. (7.15) into the modal form of Eq. (7.17). Solution. According to Eqs. (7.37) and (7.38), we need first to find the eigenvectors and eigenvalues of the Ac matrix. We take the eigenvectors to be     t11 t12 and . t21 t22 The equations using the eigenvector on the left are      −7 −12 t11 t11 =p , 1 0 t21 t21

(7.39a)

−7t11 − 12t21 = pt11 ,

(7.39b)

t11 = pt21 .

(7.39c)

Substituting Eq. (7.39c) into Eq. (7.39b) results in −7pt21 − 12t21 = p2 t21 ,

(7.40a)

p t21 + 7pt21 + 12t21 = 0,

(7.40b)

p2 + 7p + 12 = 0,

(7.40c)

2

p = −3, −4.

(7.40d)

We have found (again!) that the eigenvalues (poles) are −3 and −4; furthermore, Eq. (7.39c) tells us that the two eigenvectors are     −3t22 −4t21 and , t21 t22 4 This algorithm is part of MATLAB and all other well-known computer-aided design packages. It is

carefully documented in the software package LAPACK (Anderson et al., 1999). See also Strang (1988).

434 Chapter 7 State-Space Design where t21 and t22 are arbitrary nonzero scale factors. We want to select the two scale factors such that both elements of Bm in Eq. (7.17a) are unity. The equation for Bm in terms of Bc is TBm = Bc , and its solution is t21 = −1 and t22 = 1. Therefore, the transformation matrix and its inverse5 are     4 −3 1 3 −1 . (7.41) T= , T = 1 4 −1 1 Elementary matrix multiplication shows that, using T as defined by Eq. (7.41), the matrices of Eqs. (7.15) and (7.17) are related as follows: Am = T−1 Ac T,

Bm = T−1 Bc ,

Cm = Cc T,

Dm = Dc .

(7.42)

These computations can be carried out by using the following MATLAB statements T = [4 −3; −1 1]; Am = inv(T)*Ac*T; Bm = inv(T)*Bc; Cm = Cc*T; Dm = Dc; The next example has five state-variables and, in state-variable form, is too complicated for hand calculations. However, it is a good example for illustrating the use of computer software designed for the purpose. The model we will use is based on a physical state after amplitude and time scaling have been done.

EXAMPLE 7.11

Using MATLAB to Find Poles and Zeros of Tape-Drive System Find the eigenvalues of the system matrix described below for the tape-drive control (see Fig. 3.50). Also, compute the transformation of the equations of the tape drive in their given form to modal canonical form. The system matrices are ⎤ ⎡ ⎤ ⎡ 0 2 0 0 0 0 ⎥ ⎢0⎥ ⎢−0.1 −0.35 0.1 0.1 0.75 ⎥ ⎢ ⎥ ⎢ ⎢ ⎥ 0 0 2 0 ⎥ (7.43) F=⎢ ⎢ 0 ⎥ , G = ⎢0⎥ , ⎣ 0.4 ⎣0⎦ ⎦ 0.4 −0.4 −1.4 0 0 −0.03 0 0 −1 1  H2 = 0.0 0.0 1.0 0.0 0.0 Servomotor position output,  H3 = 0.5 0.0 0.5 0.0 0.0 Position at read/write head as output,  HT = −0.2 −0.2 0.2 0.2 0.0 Tension output, J = 0.0. 5 To find the inverse of a 2 × 2 matrix, you need only interchange the elements subscripted “11” and “22,”

change the signs of the “12” and the “21” elements, and divide by the determinant [= 1 in Eq. (7.41)].

7.4 Analysis of the State Equations The state vector is defined as ⎡ x1 (tape position at capstan) ⎢ ω1 (speed of the drive wheel) ⎢ x =⎢ ⎢ x3 (position of the tape at the head) ⎣ ω2 (output speed) i (current into capstan motor)

435

⎤ ⎥ ⎥ ⎥. ⎥ ⎦

The matrix H3 corresponds to making x3 (the position of the tape over the read/write head) the output, and the matrix HT corresponds to making tension the output. Solution. To compute the eigenvalues by using MATLAB, we write P=eig(F), which results in

⎡ ⎢ ⎢ P=⎢ ⎢ ⎣

MATLAB canon

−0.6371 + 0.6669i −0.6371 − 0.6669i 0.0000 −0.5075 −0.9683

⎤ ⎥ ⎥ ⎥. ⎥ ⎦

Notice that the system has all poles in the left half-plane (LHP) except for one pole at the origin. This means that a step input will result in a ramp output, so we conclude the system has Type 1 behavior. To transform to modal form, we use the MATLAB function canon: sysG = ss(F,G,H3,J) [sysGm, TI] =canon(sysG, ’modal’) [Am,Bm,Cm,Dm]=ssdata(sysGm) The result of this calculation is ⎡ −0.6371 0.6669 ⎢ −0.6669 −0.6371 ⎢ 0.0000 Am = Am = ⎢ ⎢ 0.0000 ⎣ 0.0000 0.0000 0.0000 0.0000

0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 0.0000 0.0000 −0.5075 0.0000

0.0000 0.0000 0.0000 0.0000 −0.9683

⎤ ⎥ ⎥ ⎥. ⎥ ⎦

Notice that the complex poles appear in the 2 × 2 block in the upper left corner of Am , and the real poles fall on the main diagonal of this matrix. The rest of the calculations from canon are ⎡ ⎤ 0.4785 ⎢ −0.6274 ⎥ ⎢ ⎥ ⎥ Bm = Bm = ⎢ ⎢ −1.0150 ⎥ , ⎣ −3.5980 ⎦ 4.9133 Cm = Cm = [ 1.2569 Dm = Dm = 0,

−1.0817

−2.8284

1.8233

0.4903 ],

436 Chapter 7 State-Space Design

TI = T

MATLAB inv

⎡ −1

⎢ ⎢ =⎢ ⎢ ⎣

−0.3439 −0.3264 0.3439 0.1847 −0.7291 −0.1847 −0.1844 −1.3533 −0.1692 0.3353 −2.3627 −0.3353 −0.0017 0.2077 0.0017

0.7741 0.4785 0.0969 −0.6247 −0.3383 −1.0150 −1.0161 −3.5980 0.0561 4.9133

⎤ ⎥ ⎥ ⎥. ⎥ ⎦

It happens that canon was written to compute the inverse of the transformation we are working with (as you can see from TI in the previous equation), so we need to invert our MATLAB results. The inverse is calculated from T = inv(T I) and results in

⎡

⎢ ⎢ T=T=⎢ ⎢ ⎣

0.3805 0.8697 −2.8284 −0.4112 −0.1502 0.0000 2.1334 −3.0330 −2.8284 0.3317 1.6776 0.0000 0.0130 −0.0114 −0.0000

⎤ 1.3406 0.4714 −0.3402 −0.2282 ⎥ ⎥ 2.3060 0.5093 ⎥ ⎥. −0.5851 −0.2466 ⎦ 0.0207 0.2160

The eigenvectors computed with [V,P]=eig(F) are V=V ⎡ ⎢ ⎢ =⎢ ⎢ ⎣

−0.1168 + 0.1925i −0.0270 − 0.1003i 0.8797 −0.2802 + 0.2933i 0.0040 + 0.0010i

−0.1168 − 0.1925i −0.0270 + 0.1003i 0.8797 −0.2802 − 0.2933i 0.0040 − 0.0010i

−0.7071 −0.0000 −0.7071 −0.0000 0.0000

0.4871 −0.1236 0.8379 −0.2126 0.0075

0.5887 −0.2850 0.6360 −0.3079 0.2697

⎤ ⎥ ⎥ ⎥. ⎥ ⎦

Notice that the first two columns of the real transformation T are composed of the real and the imaginary parts of the first eigenvector in the first column of V. It is this step that causes the complex roots to appear in the 2 × 2 block in the upper left of the Am matrix. The vectors in V are normalized to unit length, which results in nonnormalized values in Bm and Cm . If we found it desirable to do so, we could readily find further transformations to make each element of Bm equal 1 or to interchange the order in which the poles appear.

7.4.2 Dynamic Response from the State Equations Having considered the structure of the state-variable equations, we now turn to finding the dynamic response from the state description and to the relationships between the state description and our earlier discussion in Chapter 6 of the frequency response and poles and zeros. Let us begin with the general equations of state given by Eqs. (7.21a) and (7.21b), and consider the problem in the frequency domain. Taking the Laplace transform of x˙ = Fx + Gu, (7.44) we obtain sX(s) − x(0) = FX(s) + GU(s),

(7.45)

7.4 Analysis of the State Equations

437

which is now an algebraic equation. If we collect the terms involving X(s) on the left side of Eq. (7.45), keeping in mind that in matrix multiplication order is very important, we find that6 (sI − F)X(s) = GU(s) + x(0). If we premultiply both sides by the inverse of (sI − F), then X(s) = (sI − F)−1 GU(s) + (sI − F)−1 x(0).

(7.46)

The output of the system is Y (s) = HX(s) + JU(s),

(7.47a)

= H(sI − F)−1 GU(s) + H(sI − F)−1 x(0) + JU(s). Transfer function from state equations

This equation expresses the output response to both an initial condition and an external forcing input. The coefficient of the external input is the transfer function of the system, which in this case is given by G(s) =

EXAMPLE 7.12

(7.47b)

Y (s) = H(sI − F)−1 G + J. U(s)

(7.48)

Thermal System Transfer Function from the State Description Use Eq. (7.48) to find the transfer function of the thermal system described by Eqs. (7.15a) and (7.15b). Solution. The state-variable description matrices of the system are   −7 −12 F= , 1 0   1 G= , 0 H=[ 1

2 ],

J = 0.

To compute the transfer function according to Eq. (7.48), we form   s + 7 12 sI − F = −1 s and compute



(sI − F)−1

 s −12 1 s+7 . = s(s + 7) + 12

(7.49)

6 The identity matrix I is a matrix of ones on the main diagonal and zeros everywhere else; therefore,

Ix = x.

438 Chapter 7 State-Space Design We then substitute Eq. (7.49) into Eq. (7.48) to get    s −12 1 [ 1 2 ] 1 s+7 0 G(s) = s(s + 7) + 12   s [ 1 2 ] 1 = s(s + 7) + 12 (s + 2) = . (s + 3)(s + 4)

(7.50)

(7.51) (7.52)

The results can also be found using the MATLAB statements, [num,den] = ss2tf(F,G,H,J) and yield num = [0 1 2] and den = [1 7 12], which agrees with hand calculations. Because Eq. (7.48) expresses the transfer function in terms of the general statespace descriptor matrices F, G, H, and J, we are able to express poles and zeros in terms of these matrices. We saw earlier that by transforming the state matrices to diagonal form, the poles appear as the eigenvalues on the main diagonal of the F matrix. We now take a systems theory point of view to look at the poles and zeros as they are involved in the transient response of a system. As we saw in Chapter 3, a pole of the transfer function G(s) is a value of generalized frequency s such that, if s = pi , then the system can respond to an initial condition as Ki epi t , with no forcing function. In this context, pi is called a natural frequency or natural mode of the system. If we take the state-space equations (7.21a and 7.21b) and set the forcing function u to zero, we have x˙ = Fx.

(7.53)

If we assume some (as yet unknown) initial condition x(0) = x0

(7.54)

and that the entire state motion behaves according to the same natural frequency, then the state can be written as x(t) = epi t x0 . It follows from Eq. (7.53) that x˙ (t) = pi epi t x0 = Fx = Fepi t x0 ,

(7.55)

Fx0 = pi x0 .

(7.56)

(pi I − F)x0 = 0.

(7.57)

or We can rewrite Eq. (7.56) as

Transfer function poles from state equations

Equations (7.56) and (7.57) constitute the eigenvector/eigenvalue problem we saw in Eq. (7.38) with eigenvalues pi and, in this case, eigenvectors x0 of the matrix F. If we are just interested in the eigenvalues, we can use the fact that for a nonzero x0 , Eq. (7.57) has a solution if and only if det(pi I − F) = 0.

(7.58)

7.4 Analysis of the State Equations

439

These equations show again that the poles of the transfer function are the eigenvalues of the system matrix F. The determinant equation (7.58) is a polynomial in the eigenvalues pi known as the characteristic equation. In Example 7.10 we computed the eigenvalues and eigenvectors of a particular matrix in control canonical form. As an alternative computation for the poles of that system, we could solve the characteristic equation (7.58). For the system described by Eqs. (7.15a) and (7.15b), we can find the poles from Eq. (7.58) by solving  det

det(sI − F) = 0,  s + 7 12 = 0, −1 s

(7.59a) (7.59b)

s(s + 7) + 12 = (s + 3)(s + 4) = 0.

(7.59c)

This confirms again that the poles of the system are the eigenvalues of F. We can also determine the zeros of a system from the state-variable description matrices F, G, H, and J using a systems theory point of view. From this perspective, a zero is a value of generalized frequency s such that the system can have a nonzero input and state and yet have an output of zero. If the input is exponential at the zero frequency zi , given by (7.60) u(t) = u0 ezi t , then the output is identically zero: y(t) ≡ 0.

(7.61)

The state-space description of Eqs. (7.60) and (7.61) would be u = u0 ezi t ,

x(t) = x0 ezi t ,

y(t) ≡ 0.

(7.62)

Thus x˙ = zi ezi t x0 = Fezi t x0 + Gu0 ezi t , 

or [zi I − F

−G

x0 u0

(7.63)

 =0

(7.64)

and y = Hx + Ju = Hezi t x0 + Ju0 ezi t ≡ 0 Combining Eqs. (7.64) and (7.65), we get      x0 0 zi I − F −G = . u0 H J 0

(7.65)

(7.66)

Transfer function zeros from state equations

From Eq. (7.66) we can conclude that a zero of the state-space system is a value of zi where Eq. (7.66) has a nontrivial solution. With one input and one output, the matrix is square, and a solution to Eq. (7.66) is equivalent to a solution to   zi I − F −G det = 0. (7.67) H J

EXAMPLE 7.13

Zeros for the Thermal System from a State Description Compute the zero(s) of the thermal system described by Eq. (7.15).

440 Chapter 7 State-Space Design Solution. We use Eq. (7.67) to compute the zeros: ⎡ ⎤ s + 7 12 −1 s 0 ⎦ = 0, det ⎣ −1 1 2 0 −2 − s = 0, s = −2. Note that this result agrees with the zero of the transfer function given by Eq. (7.12). The result can also be found using the following MATLAB statements: sysG = ss(Ac,Bc,Cc,Dc); z = tzero(sysG) and yields z = −2.0. Equation (7.58) for the characteristic equation and Eq. (7.67) for the zeros polynomial can be combined to express the transfer function in a compact form from state-description matrices as   sI − F −G det H J G(s) = . (7.68) det(sI − F) (See Appendix WE for more details.) While Eq. (7.68) is a compact formula for theoretical studies, it is very sensitive to numerical errors. A numerically stable algorithm for computing the transfer function is described in Emami-Naeini and Van Dooren (1982). Given the transfer function, we can compute the frequency response as G(jω), and as discussed earlier, we can use Eqs. (7.57) and (7.66) to find the poles and zeros, upon which the transient response depends, as we saw in Chapter 3.

EXAMPLE 7.14

Analysis of the State Equations of a Tape Drive Compute the poles, zeros, and transfer function for the equations of the tape-drive servomechanism given in Example 7.11.

MATLAB ss2tf

Solution. There are two different ways to compute the answer to this problem. The most direct is to use the MATLAB function ss2tf (state-space to transfer function), which will give the numerator and denominator polynomials directly. This function permits multiple inputs and outputs; the fifth argument of the function tells which input is to be used. We have only one input here but must still provide the argument. The computation of the transfer function from motor-current input to the servomotor position output is [N2, D2] = ss2tf(F, G, H2, J, 1) which results in N2 = [ 0

0

D2 = [ 1.0000

0.0000 2.7500

0

0.6000 3.2225

1.2000 ], 1.8815

0.4180

−0.0000 ].

7.4 Analysis of the State Equations

441

Similarly, for the position at the read/write head, the transfer function polynomials are computed by [N3, D3] = ss2tf(F, G, H3, J, 1), which results in N3 = [ 0

−0.0000

D3 = [ 1.0000

−0.0000

2.7500

3.2225

0.7500 1.8815

1.3500

1.2000 ],

0.4180

−0.0000 ].

Finally, the transfer function to tension is [NT, DT] = ss2tf(F, G, HT, J, 1) producing NT = [ 0

−0.0000

DT = [ 1.0000 MATLAB roots

−0.1500

2.7500

3.2225

−0.4500 1.8815

−0.3000 0.4180

0.0000 ], −0.0000 ].

It is interesting to check to see whether the poles and zeros determined this way agree with those found by other means. For a polynomial, we use the function roots: ⎡ ⎤ −0.6371 + 0.6669i ⎢ −0.6371 − 0.6669i ⎥ ⎢ ⎥ ⎥. roots(D3) = ⎢ ⎢ −0.9683 ⎥ ⎣ −0.5075 ⎦ 0.0000 Checking with Example 7.11, we confirm that they agree. How about the zeros? We can find these by finding the roots of the numerator polynomial. We compute the roots of the polynomial N3: ⎡ ⎤ −1.6777 × 107 ⎢ 1.6777 × 107 ⎥ ⎥ roots(N3) = ⎢ ⎣ −0.9000 + 0.8888i ⎦ . −0.9000 − 0.8888i Here we notice that roots are given with a magnitude of 107 , which seems inconsistent with the values given for the polynomial. The problem is that MATLAB has used the very small leading terms in the polynomial as real values and thereby introduced extraneous roots that are for all practical purposes at infinity. The true zeros are found by truncating the polynomial to the significant values using the statement N3R = N3(4 : 6) to get N3R = [ 0.7499 1.3499 1.200 ],   −0.9000 + 0.8888i roots(N3R) = . −0.9000 − 0.8888i The other approach is to compute the poles and zeros separately and, if desired, combine these into a transfer function. The poles were computed with eig in Example 7.11 and are

442 Chapter 7 State-Space Design ⎡ ⎢ ⎢ P=⎢ ⎢ ⎣ MATLAB tzero

−0.6371 + 0.6669i −0.6371 − 0.6669i 0.0000 −0.5075 −0.9683

⎤ ⎥ ⎥ ⎥. ⎥ ⎦

The zeros can be computed by the equivalent of Eq. (7.66) with the function tzero (transmission zeros). The zeros depend on which output is being used, of course, and are respectively given below. For the position of the tape at the servomotor as the output, the statement sysG2 = ss(F, G, H2, J) ZER2 = tzero(sysG2) yields ZER2 = −2.0000. For the position of the tape over the read/write head as the output, we use the statement sysG3 = ss(F, G, H3, J) ZER3 = tzero(sysG3)   −0.9000 + 0.8888i ZER3 = . −0.9000 − 0.8888i We note that these results agree with the values previously computed from the numerator polynomial N3. Finally, for the tension as output, we use sysGT = ss(F, G, HT, J) ZERT = tzero(sysGT) to get

⎡

⎤ 0 ZERT = ⎣ −1.9999 ⎦ . −1.0000 From these results we can write down, for example, the transfer function to position x3 as G(s) =

X3 (s) E1 (s)

0.75s2 + 1.35s + 1.2 + 2.75s4 + 3.22s3 + 1.88s2 + 0.418s 0.75(s + 0.9 ± 0.8888j) = . s(s + 0.507)(s + 0.968)(s + 0.637 ± 0.667j)

=

s5

(7.69)

7.5 Control-Law Design for Full-State Feedback One of the attractive features of the state-space design method is that it consists of a sequence of independent steps, as mentioned in the chapter overview. The first step,

7.5 Control-Law Design for Full-State Feedback

Estimator/observer

The control law and the estimator together form the compensation

443

discussed in Section 7.5.1, is to determine the control. The purpose of the control law is to allow us to assign a set of pole locations for the closed-loop system that will correspond to satisfactory dynamic response in terms of rise time and other measures of transient response. In Section 7.5.2 we will show how to introduce the reference input with full-state feedback, and in Section 7.6 we will describe the process of finding the poles for good design. The second step—necessary if the full state is not available—is to design an estimator (sometimes called an observer), which computes an estimate of the entire state vector when provided with the measurements of the system indicated by Eq. (7.21b). We will examine estimator design in Section 7.7. The third step consists of combining the control law and the estimator. Figure 7.11 shows how the control law and the estimator fit together and how the combination takes the place of what we have been previously referring to as compensation. At this stage, the control-law calculations are based on the estimated state rather than the actual state. In Section 7.8 we will show that this substitution is reasonable, and also that using the combined control law and estimator results in closed-loop pole locations that are the same as those determined when designing the control and estimator separately. The fourth and final step of state-space design is to introduce the reference input in such a way that the plant output will track external commands with acceptable rise-time, overshoot, and settling-time values. At this point in the design, all the closed-loop poles have been selected, and the designer is concerned with the zeros of the overall transfer function. Figure 7.11 shows the command input r introduced in the same relative position as was done with the transform design methods; however, in Section 7.9 we will show how to introduce the reference at another location, resulting in different zeros and (usually) superior control.

7.5.1 Finding the Control Law Control law

Figure 7.11 Schematic diagram of state-space design elements

The first step in the state-space design method, as mentioned earlier, is to find the control law as feedback of a linear combination of the state-variables—that is,

Plant u

x  Fx  Gu

x

Control law

ˆ x

K Matrix of constants

Y

H

 Estimator

State vector estimate

Compensation






R

444 Chapter 7 State-Space Design ⎡ u = −Kx = −[ K1

K2

⎢ ⎢ · · · Kn ] ⎢ ⎣

x1 x2 .. .

⎤ ⎥ ⎥ ⎥. ⎦

(7.70)

xn We assume for feedback purposes that all the elements of the state vector are at our disposal. In practice, of course, this would usually be a ridiculous assumption; moreover, a well-trained control designer knows that other design methods do not require so many sensors. The assumption that all state-variables are available merely allows us to proceed with this first step. Equation (7.70) tells us that the system has a constant matrix in the state-vector feedback path, as shown in Fig. 7.12. For an nth-order system, there will be n feedback gains, K1 , . . . , Kn , and because there are n roots of the system, it is possible that there are enough degrees of freedom to select arbitrarily any desired root location by choosing the proper values of Ki . This freedom contrasts sharply with root-locus design, in which we have only one parameter and the closed-loop poles are restricted to the locus. Substituting the feedback law given by Eq. (7.70) into the system described by Eq. (7.21a) yields x˙ = Fx − GKx. (7.71) Control characteristic equation

The characteristic equation of this closed-loop system is det[sI − (F − GK)] = 0.

(7.72)

When evaluated, this yields an nth-order polynomial in s containing the gains K1 , . . . , Kn . The control-law design then consists of picking the gains K so that the roots of Eq. (7.72) are in desirable locations. Selecting desirable root locations is an inexact science that may require some iteration by the designer. Issues in their selection are considered in Examples 7.15 to 7.17 as well as in Section 7.6. For now, we assume that the desired locations are known, say, s = s1 , s2 , . . . , sn . Then the corresponding desired (control) characteristic equation is αc (s) = (s − s1 )(s − s2 ) . . . (s − sn ) = 0.

(7.73)

Hence the required elements of K are obtained by matching coefficients in Eqs. (7.72) and (7.73). This forces the system’s characteristic equation to be identical to the desired characteristic equation and the closed-loop poles to be placed at the desired locations. Figure 7.12 Assumed system for control-law design

u

x  Fx  Gu

u  Kx

x

H

Y

7.5 Control-Law Design for Full-State Feedback

EXAMPLE 7.15

445

Control Law for a Pendulum Suppose you have a pendulum with frequency ω0 and a state-space description given by        0 1 x˙ 1 0 x1 = + u. (7.74) x˙ 2 x2 1 −ω02 0 Find the control law that places the closed-loop poles of the system so that they are both at −2ω0 . In other words, you wish to double the natural frequency and increase the damping ratio ζ from 0 to 1. Solution. From Eq. (7.73) we find that αc (s) = (s + 2ω0 )2 =s

2

(7.75a)

+ 4ω0 s + 4ω02 .

Equation (7.72) tells us that       0 1 s 0 0 − [ K1 det[sI − (F − GK)] = det − 1 0 s −ω02 0 or s2 + K2 s + ω02 + K1 = 0.

(7.75b)  K2 ]

,

(7.76)

Equating the coefficients with like powers of s in Eqs. (7.75b) and (7.76) yields the system of equations K2 = 4ω0 , ω02

+ K1 = 4ω02 ,

and therefore, K1 = 3ω02 , K2 = 4ω0 . More concisely, the control law is K = [ K1

K2 ] = [ 3ω02

4ω0 ].

Figure 7.13 shows the response of the closed-loop system to the initial conditions x1 = 1.0, x2 = 0.0, and ω0 = 1. It shows a very well damped response, as would be expected from having two roots at s = −2. The MATLAB command impulse was used to generate the plot. Calculating the gains by using the technique illustrated in Example 7.15 becomes rather tedious when the order of the system is higher than 3. There are, however, special “canonical” forms of the state-variable equations for which the algebra for finding the gains is especially simple. One such canonical form that is useful in control law design is the control canonical form. Consider the third-order system7 ... y + a1 y¨ + a2 y˙ + a3 y = b1 u¨ + b2 u˙ + b3 u, (7.77) 7 This development is exactly the same for higher-order systems.

446 Chapter 7 State-Space Design Figure 7.13 Impulse response of the undamped oscillator with full-state feedback (ω0 = 1)

1.0 0.8

x1

0.6 Amplitude

0.4 0.2 u/4 0 0.2

x2

0.4 0.6 0.8

0

1

2

3 4 Time (sec)

5

6

7

which corresponds to the transfer function G(s) =

b 1 s 2 + b2 s + b 3 Y (s) b(s) = 3 . = 2 U(s) s + a1 s + a 2 s + a 3 a(s)

(7.78)

Suppose we introduce an auxiliary variable (referred to as the partial state) ξ , which relates a(s) and b(s) as shown in Fig. 7.14(a). The transfer function from U to ξ is

or

ξ(s) 1 = , U(s) a(s)

(7.79)

... ξ + a1 ξ¨ + a2 ξ˙ + a3 ξ = u.

(7.80)

It is easy to draw a block diagram corresponding to Eq. (7.80) if we rearrange the equation as follows: ... ξ = −a1 ξ¨ − a2 ξ˙ − a3 ξ + u. (7.81) The summation is indicated in Fig. 7.14(b), where each ξ on the right-hand side is ... obtained by sequential integration of ξ . To form the output, we go back to Fig. 7.14(a) and note that Y (s) = b(s)ξ(s), (7.82) which means that

y = b1 ξ¨ + b2 ξ˙ + b3 ξ .

(7.83)

We again pick off the outputs of the integrators, multiply them by {bi }’s, and form the right-hand side of Eq. (7.77) by using a summer to yield the output as shown in Fig. 7.14(c). In this case, all the feedback loops return to the point of the application of the input, or “control” variable, and hence the form is referred to as the control canonical form. Reduction of the structure by Mason’s rule or by elementary block diagram operations verifies that this structure has the transfer function given by G(s). Taking the state as the outputs of the three integrators numbered, by convention, from the left, namely (7.84) x1 = ξ¨1 , x2 = ξ˙ , x3 = ξ ,

7.5 Control-Law Design for Full-State Feedback Figure 7.14 Derivation of control canonical form



1 a(s)

U

447

Y

b(s)

(a)

U

 








1 s







1 s

1 s



a1 a2 a3 (b)

b1

 b2

U

 






1 s





 x1c



1 s

x2c

1 s

  x3c




Y



b3

a1 a2 a3 (c)

we obtain ... x˙ 1 = ξ = −a1 x1 − a2 x2 − a3 x3 + u, x˙ 2 = x1 , x˙ 3 = x2 .

(7.85)

We may now write the matrices describing the control canonical form in general: ⎡ ⎡ ⎤ ⎤ −a1 −a2 · · · · · · −an 1 ⎢ 1 ⎢0 ⎥ ⎥ 0 · · · · · · 0 ⎢ ⎢ ⎥ ⎥ ⎢ ⎢ ⎥ 1 0 ··· 0 ⎥ (7.86a) Gc = ⎢0⎥ , Fc = ⎢ 0 ⎥, ⎢ .. ⎢ .. ⎥ ⎥ . . ⎣ . ⎣.⎦ . 0 ⎦ 

Hc = b1 Companion form matrix

0 b2

0

···

···

· · · bn ,

1



0

0 Jc = 0.

(7.86b)

The special structure of this system matrix is referred to as the upper companion form because the characteristic equation is a(s) = sn + a1 sn−1 + a2 sn−2 · · · + an and

448 Chapter 7 State-Space Design the coefficients of this monic “companion” polynomial are the elements in the first row of Fc . If we now form the closed-loop system matrix Fc − Gc Kc , we find that ⎡ ⎤ −a1 − K1 −a2 − K2 · · · · · · −an − Kn ⎢ ⎥ 1 0 ··· ··· 0 ⎢ ⎥ ⎥ ⎢ 0 1 0 ··· 0 (7.87) Fc − Gc Kc = ⎢ ⎥. ⎥ ⎢ .. . .. .. ⎦ ⎣ . . 0

0

···

1

0

By visually comparing Eqs. (7.86a) and (7.87), we see that the closed-loop characteristic equation is sn + (a1 + K1 )sn−1 + (a2 + K2 )sn−2 + · · · + (an + Kn ) = 0.

(7.88)

Therefore, if the desired pole locations result in the characteristic equation given by αc (s) = sn + α1 sn−1 + α2 sn−2 + · · · + αn = 0,

(7.89)

then the necessary feedback gains can be found by equating the coefficients in Eqs. (7.88) and (7.89): K1 = −a1 + α1 , K2 = −a2 + α2 , . . . , Kn = −an + αn .

Ackermann’s formula for pole placement

(7.90)

We now have the basis for a design procedure. Given a system of order n described by an arbitrary (F, G) and given a desired nth-order monic characteristic polynomial αc (s), we (1) transform (F, G) to control canonical form (Fc , Gc ) by changing the state x = Tz, and we (2) solve for the control gains by inspection using Eq. (7.90) to give the control law u = −Kc z. Because this gain is for the state in the control form, we must (3) transform the gain back to the original state to get K = Kc T−1 . An alternative to this transformation method is given by Ackermann’s formula (1972), which organizes the three-step process of converting to (Fc , Gc ), solving for the gains, and converting back again into the very compact form K=[ 0

··· 0

1 ]C −1 αc (F),

(7.91)

such that C=[ G

FG

F2 G

· · · Fn−1 G ],

(7.92)

where C is the controllability matrix we saw in Section 7.4, n gives the order of the system and the number of state-variables, and αc (F) is a matrix defined as αc (F) = Fn + α1 Fn−1 + α2 Fn−2 + · · · + αn I,

(7.93)

where the αi are the coefficients of the desired characteristic polynomial Eq. (7.89). Note that Eq. (7.93) is a matrix equation. Refer to Appendix WG for the derivation of Ackermann’s formula.

EXAMPLE 7.16

Ackermann’s Formula for Undamped Oscillator (a) Use Ackermann’s formula to solve for the gains for the undamped oscillator of Example 7.15. (b) Verify the calculations with MATLAB for ω0 = 1.

7.5 Control-Law Design for Full-State Feedback

449

Solution (a) The desired characteristic equation is αc (s) = (s+2ω0 )2 . Therefore, the desired characteristic polynomial coefficients, α1 = 4ω0 ,

α2 = 4ω02 ,

are substituted into Eq. (7.93) and the result is      0 1 −ω02 0 1 2 + 4ω αc (F) = + 4ω 0 0 0 −ω02 0 0 −ω02   2 4ω0 3ω0 = . −4ω03 3ω02 The controllability matrix is



C=[ G

FG ] =

which yields

0 1

1 0

0 1

K = [ K1 =[ 0

K2 ]  0 1 ] 1

, (7.94a) (7.94b)

 ,

 0 1 . 1 0 Finally, we substitute Eqs. (7.95) and (7.94a) into Eq. (7.91) to get C −1 =





1 0



3ω02 −4ω03

4ω0 3ω02

(7.95)

 .

Therefore K = [ 3ω02 4ω0 ], which is the same result we obtained previously. (b) The MATLAB statements wo = 1; F = [0 1;−wo*wo 0]; G = [0;1]; pc = [−2*wo;−2*wo]; K = acker(F,G,pc) yield K = [ 3

MATLAB acker, place

4 ], which agrees with hand calculations.

As was mentioned earlier, computation of the controllability matrix has very poor numerical accuracy, and this carries over to Ackermann’s formula. Equation (7.91), implemented in MATLAB with the function acker, can be used for the design of singleinput-single-output (SISO) systems with a small (≤10) number of state-variables. For more complex cases a more reliable formula is available, implemented in MATLAB with place. A modest limitation on place is that, because it is based on assigning closed-loop eigenvectors, none of the desired closed-loop poles may be repeated; that is, the poles must be distinct,8 a requirement that does not apply to acker. 8 One may get around this restriction by moving the repeated poles by very small amounts to make them distinct.

450 Chapter 7 State-Space Design

An example of weak controllability

The fact that we can shift the poles of a system by state feedback to any desired location is a rather remarkable result. The development in this section reveals that this shift is possible if we can transform (F, G) to the control form (Fc , Gc ), which in turn is possible if the system is controllable. In rare instances the system may be uncontrollable, in which case no possible control will yield arbitrary pole locations. Uncontrollable systems have certain modes, or subsystems, that are unaffected by the control. This usually means that parts of the system are physically disconnected from the input. For example, in modal canonical form for a system with distinct poles, one of the modal state-variables is not connected to the input if there is a zero entry in the Bm matrix. A good physical understanding of the system being controlled would prevent any attempt to design a controller for an uncontrollable system. As we saw earlier, there are algebraic tests for controllability; however, no mathematical test can replace the control engineer’s understanding of the physical system. Often the physical situation is such that every mode is controllable to some degree, and, while the mathematical tests indicate the system is controllable, certain modes are so weakly controllable that designs to control them are virtually useless. Airplane control is a good example of weak controllability of certain modes. Pitch plane motion xp is primarily affected by the elevator δe and weakly affected by rolling motion xr . Rolling motion is essentially affected only by the ailerons δa . The state-space description of these relationships is         Fp ε xp Gp 0 δe x˙ p = + , (7.96) x˙ r xr δa 0 Fr 0 Gr where the matrix of small numbers ε represents the weak coupling from rolling motion to pitching motion. A mathematical test of controllability for this system would conclude that pitch plane motion (and therefore altitude) is controllable by the ailerons as well as by the elevator! However, it is impractical to attempt to control an airplane’s altitude by rolling the aircraft with the ailerons. Another example will illustrate some of the properties of pole placement by state feedback and the effects of loss of controllability on the process.

EXAMPLE 7.17

How Zero Location Can Affect the Control Law A specific thermal system is described by Eq. (7.35a) in observer canonical form with a zero at s = z0 . (a) Find the state feedback gains necessary forplacing the poles of this system at the roots of s2 + 2ζ ωn s + ωn2 (i.e., at −ζ ωn ± jωn 1 − ζ 2 ). (b) Repeat the computation with MATLAB, using the parameter values z0 = 2, ζ = 0.5, and ωn = 2 rad/sec. Solution (a) The state description matrices  are  −7 1 Ao = , −12 0

 Bo =

 1 , −z0

Do = 0. Co = [1 0], First we substitute these matrices into Eq. (7.72) to get the closed-loop characteristic equation in terms of the unknown gains and the zero position: s2 + (7 + K1 − z0 K2 )s + 12 − K2 (7z0 + 12) − K1 z0 = 0.

7.5 Control-Law Design for Full-State Feedback

451

Next we equate this equation to the desired characteristic equation to get the equations K1 − z0 K2 = 2ζ ωn − 7, −z0 K1 − (7z0 + 12)K2 = ωn2 − 12. The solutions to these equations are K1 =

z0 (14ζ ωn − 37 − ωn2 ) + 12(2ζ ωn − 7) , (z0 + 3)(z0 + 4)

K2 =

z0 (7 − 2ζ ωn ) + 12 − ωn2 . (z0 + 3)(z0 + 4)

(b) The following MATLAB statements can be used to find the solution: Ao = [−7 1;−12 0]; zo = 2; Bo = [1;−zo]; pc = roots([1 2 4]); K = place(Ao,Bo,pc) These statements yield K=[−3.80 0.60], which agrees with the hand calculations. If the zero were close to one of the open-loop poles, say z0 = −2.99, then we find K=[2052.5 −688.1]. Two important observations should be made from this example. The first is that the gains grow as the zero z0 approaches either −3 or −4, the values where this system loses controllability. In other words, as controllability is almost lost, the control gains become very large. The system has to work harder and harder to achieve control as controllability slips away. The second important observation illustrated by the example is that both K1 and K2 grow as the desired closed-loop bandwidth given by ωn is increased. From this, we conclude that To move the poles a long way requires large gains. These observations lead us to a discussion of how we might go about selecting desired pole locations in general. Before we begin that topic, we will complete the design with full-state feedback by showing how the reference input might be applied to such a system and what the resulting response characteristics are.

7.5.2 Introducing the Reference Input with Full-State Feedback Thus far, the control has been given by Eq. (7.70), or u = −Kx. In order to study the transient response of the pole-placement designs to input commands, it is necessary

452 Chapter 7 State-Space Design to introduce the reference input into the system. An obvious way to do this is to change the control to u = −Kx + r. However, the system will now almost surely have a nonzero steady-state error to a step input. The way to correct this problem is to compute the steady-state values of the state and the control input that will result in zero output error and then force them to take these values. If the desired final values of the state and the control input are xss and uss respectively, then the new control formula should be (7.97) u = uss − K(x − xss ), so that when x = xss (no error), u = uss . To pick the correct final values, we must solve the equations so that the system will have zero steady-state error to any constant input. The system differential equations are the standard ones: x˙ = Fx + Gu, (7.98a) y = Hx + Ju. In the constant steady state, Eqs. (7.98a) and (7.98b) reduce to the pair 0 = Fxss + Guss , Gain calculation for reference input

Control equation with reference input

(7.98b) (7.99a)

(7.99b) yss = Hxss + Juss . We want to solve for the values for which yss = rss for any value of rss . To do this, we make xss = Nx rss and uss = Nu rss . With these substitutions we can write Eqs. (7.99) as a matrix equation; the common factor of rss cancels out to give the equation for the gains:      0 F G Nx = . (7.100) H J Nu 1 This equation can be solved for Nx and Nu to get   −1    F G 0 Nx = . Nu H J 1 With these values, we finally have the basis for introducing the reference input so as to get zero steady-state error to a step input: u = Nu r − K(x − Nx r) (7.101a) (7.101b) = −Kx + (Nu + KNx )r. The coefficient of r in parentheses is a constant that can be computed beforehand. We ¯ so give it the symbol N, ¯ u = −Kx + Nr. (7.102) The block diagram of the system is shown in Fig. 7.15.

EXAMPLE 7.18

Introducing the Reference Input Compute the necessary gains for zero steady-state error to a step command at x1 , and plot the resulting unit step response for the oscillator in Example 7.15 with ω0 = 1. Solution. We substitute the matrices of Eq. (7.74) (with ω0 = 1 and H = [ 1 0 ] because y = x1 ) into Eq. (7.100) to get ⎤ ⎤ ⎡ ⎡  0 1 0  0 N x ⎣ −1 0 1 ⎦ (7.103) = ⎣ 0 ⎦· Nu 1 0 0 1

7.5 Control-Law Design for Full-State Feedback Figure 7.15 Block diagram for introducing the reference input with full-state feedback: (a) with state and control gains; (b) with a single composite gain

Nu

R

Plant






u

Y



N

R



Nx






x




u

Y x

K



(a)

Figure 7.16 Step response of oscillator to a reference input

Plant



K

453

(b)

1.0 x1

xss

0.8

Amplitude

0.6 0.4

uss

x2

0.2 u/4 0 0.2

0

1

2

3 4 Time (sec)

5

6

7

The solution is (x=a\b in MATLAB where a and b are the left- and right-hand side matrices respectively),   1 , Nx = 0 Nu = 1, and, for the given control law, K = [3ω02 4ω0 ] = [ 3 N¯ = Nu + KNx = 4.

4 ], (7.104)

The corresponding step response (using the MATLAB step command) is plotted in Fig. 7.16. Note that there are two equations for the control—Eqs. (7.101b) and (7.102). While these expressions are equivalent in theory, they differ in practical implementation in that Eq. (7.101b) is usually more robust to parameter errors than Eq. (7.102), particularly when the plant includes a pole at the origin and Type 1 behavior is possible. The difference is most clearly illustrated by the next example.

454 Chapter 7 State-Space Design

EXAMPLE 7.19

DC Motor Eq. (2.52)

Reference Input to a Type 1 System: DC Motor Compute the input gains necessary to introduce a reference input with zero steadystate error to a step for the DC motor of Example 5.1, which in state-variable form is described by the matrices:     0 1 0 F= , G= , 0 −1 1  H= 1 0 , J = 0. Assume that the state feedback gain is [ K1

K2 ]. Solution. If we substitute the system matrices of this example into the equation for the input gains, Eq. (7.100), we find that the solution is   1 Nx = , 0 Nu = 0, N¯ = K1 .

With these values, the expression for the control using Nx and Nu [Eq. (7.101b)] reduces to u = −K1 (x1 − r) − K2 x2 , ¯ while the one using N [Eq. (7.102)] becomes u = −K1 x1 − K2 x2 + K1 r. The block diagrams for the systems using each of the control equations are given in Fig. 7.17. When using Eq. (7.102), as shown in Fig. 7.17(b), it is necessary to multiply ¯ exactly equal to that used in the feedback. If these two the input by a gain K1 (= N) gains do not match exactly, there will be a steady-state error. On the other hand, if we use Eq. (7.101b), as shown in Fig. 7.17(a), there is only one gain to be used on the difference between the reference input and the first state, and zero steady-state error will result even if this gain is slightly in error. The system of Fig. 7.17(a) is more robust than the system of Fig. 7.17(b). With the reference input in place, the closed-loop system has input r and output y. From the state description we know that the system poles are at the eigenvalues of the closed-loop system matrix, F − GK. In order to compute the closed-loop transient response, it is necessary to know where the closed-loop zeros of the transfer function from r to y are. They are to be found by applying Eq. (7.67) to the closedloop description, which we assume has no direct path from input u to output y, so that J = 0. The zeros are values of s such that   ¯ sI − (F − GK) −NG det = 0. (7.105) H 0 We can use two elementary facts about determinants to simplify Eq. (7.105). In the ¯ which is a scalar, then the point where first place, if we divide the last column by N, the determinant is zero remains unchanged. The determinant is also not changed if we

7.6 Selection of Pole Locations for Good Design Figure 7.17 Alternative structures for introducing the reference input: (a) Eq. (7.101b); (b) Eq. (7.102)

R






K1



u








1 s

x2

1 s

x1

x2

1 s

x1

455

K2

(a)

R

N








u




1 s K2

K1 (b)

multiply the last column by K and add it to the first (block) column, with the result that the GK term is cancelled out. Thus the matrix equation for the zeros reduces to   sI − F −G det = 0. (7.106) H 0 Equation (7.106) is the same as Eq. (7.67) for the zeros of the plant before the feedback was applied. The important conclusion is that When full-state feedback is used as in Eq. (7.101b) or (7.102), the zeros remain unchanged by the feedback.

7.6 Selection of Pole Locations for Good Design

Two methods of pole selection

The first step in the pole-placement design approach is to decide on the closed-loop pole locations. When selecting pole locations, it is always useful to keep in mind that the control effort required is related to how far the open-loop poles are moved by the feedback. Furthermore, when a zero is near a pole, the system may be nearly uncontrollable, and as we saw in Section 7.5, moving such poles requires large control gains and thus a large control effort; however, the designer is able to temper the choices to take control effort into account. Therefore, a pole-placement philosophy that aims to fix only the undesirable aspects of the open-loop response and avoids either large increases in bandwidth or efforts to move poles that are near zeros will typically allow smaller gains, and thus smaller control actuators, than a philosophy that arbitrarily picks all the poles without regard to the original open-loop pole and zero locations. In this section we discuss two techniques to aid in the pole-selection process. The first approach—dominant second-order poles—deals with pole selection without explicit regard for their effect on control effort; however, the designer is able to temper

456 Chapter 7 State-Space Design the choices to take control effort into account. The second method (called optimal control, or symmetric root locus) does specifically address the issue of achieving a balance between good system response and control effort.

7.6.1 Dominant Second-Order Poles The step response corresponding to the second-order transfer function with complex poles at radius ωn and damping ratio ζ was discussed in Chapter 3. The rise time, overshoot, and settling time can be deduced directly from the pole locations. We can choose the closed-loop poles for a higher-order system as a desired pair of dominant second-order poles, and select the rest of the poles to have real parts corresponding to sufficiently damped modes, so that the system will mimic a second-order response with reasonable control effort. We also must make sure that the zeros are far enough into the LHP to avoid having any appreciable effect on the second-order behavior. A system with several lightly damped high-frequency vibration modes plus two rigidbody low-frequency modes lends itself to this philosophy. Here we can pick the low-frequency modes to achieve desired values of ωn and ζ and select the rest of the poles to increase the damping of the high-frequency modes, while holding their frequency constant in order to minimize control effort. To illustrate this design method, we obviously need a system of higher than second-order; we will use the tape drive servomotor described in Example 7.11.

EXAMPLE 7.20

Pole Placement as a Dominant Second-Order System Design the tape servomotor by the dominant second-order poles method to have no more than 5% overshoot and a rise time of no more than 4 sec. Keep the peak tension as low as possible.

MATLAB acker

Solution. From the plots of the second-order transients in Fig. 3.18, a damping ratio ζ = 0.7 will meet the overshoot requirement and, for this damping ratio, a rise time of 4 sec suggests a natural frequency of about 1/1.5. There are five poles in all, so the other three need to be placed far to the left of the dominant pair. For our purposes, “far” means the transients due to the fast poles should be over well before the transients due to the dominant poles, and we assume a factor of 4 in the respective undamped natural frequencies to be adequate. From these considerations, the desired poles are given by  pc = −0.707 + 0.707 ∗ j; −0.707 − 0.707 ∗ j; −4; −4; −4 /1.5; (7.107) With these desired poles, we can use the function acker with F and G from Example 7.11, Eq. (7.70), to find the control gains K2 = [ 8.5123 20.3457 −1.4911 −7.8821 6.1927 ]. (7.108) These are found with the following MATLAB statements: F = [0 2 0 0 0;−.1 −.35 .1 .1.75;0 0 0 2 0;.4 .4 −.4 -1.4 0;0 −.03 0 0 -1]; G = [0;0;0;0;1]; pc = [−.707+.707*j;−.707−.707*j;−4;−4;−4]/1.5; K2 = acker(F,G,pc) The step response and the corresponding tension plots for this and another design (to be discussed in Section 7.6.2) are given in Fig. 7.18 and Fig. 7.19. Notice that the rise time is approximately 4 sec and the overshoot is about 5%, as specified.

7.6 Selection of Pole Locations for Good Design x3

1.2

LQR

1.0

Tape position

Figure 7.18 Step responses of the tape servomotor designs

457

0.8

Dominant second order

0.6 0.4 0.2 0

0

2

4

6

8

10

12

Time (msec)

Figure 7.19 Tension plots for tape servomotor step responses

T

0.02

LQR

0.0

Tape tension

0.02

Dominant second order

0.04 0.06 0.08 0.10 0.12

0

2

4

6

8

10

12

Time (msec)

Because the design process is iterative, the poles we selected should be seen as only a first step, to be followed by further modifications to meet the specifications as accurately as necessary. For this example we happened to select adequate pole locations on the first try.

7.6.2 Symmetric Root Locus (SRL) LQR design

A most effective and widely used technique of linear control systems design is the optimal linear quadratic regulator (LQR). The simplified version of the LQR problem is to find the control such that the performance index

∞ J = [ρz2 (t) + u2 (t)] dt (7.109) 0

458 Chapter 7 State-Space Design is minimized for the system x˙ = Fx + Gu,

(7.110a)

z = H1 x,

(7.110b)

where ρ in Eq. (7.109) is a weighting factor of the designer’s choice. A remarkable fact is that the control law that minimizes J is given by linear-state feedback u = −Kx. Symmetric root locus

(7.111)

Here the optimal value of K is that which places the closed-loop poles at the stable roots (those in the LHP) of the symmetric root-locus (SRL) equation (Kailath, 1980) 1 + ρG0 (−s)G0 (s) = 0,

(7.112)

where G0 is the open-loop transfer function from u to z: G0 (s) =

SRL equation

Z(s) N(s) = H1 (sI − F)−1 G = . U(s) D(s)

(7.113)

Note that this is a root-locus problem as discussed in Chapter 5 with respect to the parameter ρ, which weights the relative cost of (tracking error) z2 with respect to the control effort u2 in the performance index Eq. (7.109). Note also that s and −s affect Eq. (7.112) in an identical manner; therefore, for any root s0 of Eq. (7.112), there will also be a root at −s0 . We call the resulting root locus a SRL, since the locus in the LHP will have a mirror image in the right half-plane (RHP); that is, there is symmetry with respect to the imaginary axis. We may thus choose the optimal closed-loop poles by first selecting the matrix H1 , which defines the tracking error and which the designer wishes to keep small, and then choosing ρ, which balances the importance of this tracking error against the control effort. Notice that the output we select as tracking error does not need to be the plant sensor output. That is why we call the output in Eq. (7.110) z rather than y. Selecting a set of stable poles from the solution of Eq. (7.112) results in desired closed-loop poles, which we can then use in a pole-placement calculation such as Ackermann’s formula (Eq. 7.91) to obtain K. As with all root loci for real transfer functions G0 , the locus is also symmetric with respect to the real axis; thus there is symmetry with respect to both the real and imaginary axes. We can write the SRL equation in the standard root-locus form 1+ρ

N(−s)N(s) = 0, D(−s)D(s)

(7.114)

obtain the locus poles and zeros by reflecting the open-loop poles and zeros of the transfer function from U to Z across the imaginary axis (which doubles the number of poles and zeros), and then sketch the locus. Note that the locus could be either a 0◦ or 180◦ locus, depending on the sign of G0 (−s)G0 (s) in Eq. (7.112). A quick way to determine which type of locus to use (0◦ or 180◦ ) is to pick the one that has no part on the imaginary axis. The real-axis rule of root locus plotting will reveal this right away. For the controllability assumptions we have made here, plus the assumption that all the system modes are present in the chosen output z, the optimal closed-loop system is guaranteed to be stable; thus no part of the locus can be on the imaginary axis.

7.6 Selection of Pole Locations for Good Design Figure 7.20 SRL for a first-order system

Im(s) r0

r0

r0

r0

a

EXAMPLE 7.21

459

Re(s)

a

SRL for Servo Speed Control Plot the SRL for the following servo speed control system with z = y: y˙ = −ay + u, 1 G0 (s) = . s+a

(7.115a) (7.115b)

Solution. The SRL equation [Eq. (7.112)] for this example is 1+ρ

1 = 0. (−s + a)(s + a)

(7.116)

The SRL, shown in Fig. 7.20, is a 0◦ locus. The optimal (stable) pole can be determined explicitly in this case as  s = − a2 + ρ. (7.117) Thus, the closed-loop root location that minimizes the performance index of Eq. (7.109) lies on the real axis at the distance given by Eq. (7.117) and is always to the left of the open-loop root.

EXAMPLE 7.22

SRL Design for Satellite Attitude Control Draw the SRL for the satellite system with z = y. Solution. The equations of motion are     0 1 0 x˙ = x+ u, 0 0 1 y=[ 1

0 ] x.

(7.118) (7.119)

We then calculate from Eqs. (7.118) and (7.119) so that 1 . (7.120) s2 The symmetric 180◦ loci are shown in Fig. 7.21. The MATLAB statements to generate the SRL are G0 (s) =

numGG = [1]; denGG = conv([1 0 0],[1 0 0]); sysGG = tf(numGG,denGG); rlocus(sysGG);

460 Chapter 7 State-Space Design Figure 7.21 SRL for the satellite

Im(s)

j

r0 1

1

Re(s)

j

It is interesting to note that the (stable) closed-loop poles have damping of ζ = 0.707. We would choose two stable roots for a given value of ρ, for example s = −1 ± j1 for ρ = 4.07, on the SRL and use them for pole-placement and control-law design. Choosing different values of ρ can provide us with pole  2locations that achieve varying balances between a fast response (small values of z dt) and a low control  effort (small values of u2 dt). Figure 7.22 shows the design trade-off curve for the satellite (double-integrator) plant [Eq. (7.18)] for various values of ρ ranging from 0.01 to 100. The curve has two asymptotes (dashed lines) corresponding to low (large ρ) and high (small ρ) penalty on the control usage. In practice, usually a value of ρ corresponding to a point close to the knee of the trade-off curve is chosen. This is because it provides a reasonable compromise between the use of control and the speed of response. For the satellite plant, the value of ρ = 1 corresponds to the knee of the curve. In this case the closed-loop poles have a damping ratio of ζ = 0.707! Figure 7.23 shows the associated Nyquist plot, which has a phase margin PM = 65◦ and infinite gain margin. These excellent stability properties are a general feature of LQR designs. It is also possible to locate optimal pole locations for the design of an open-loop unstable system using the SRL and LQR method.

EXAMPLE 7.23

SRL Design for an Inverted Pendulum Draw the SRL for the linearized equations of the simple inverted pendulum with ωo = 1. Take the output, z, to be the sum of twice the position plus the velocity (so as to weight or penalize both position and velocity). Solution. The equations of motion are     0 1 0 x˙ = x + u. (7.121) −1 ω02 0

7.6 Selection of Pole Locations for Good Design Figure 7.22 Design trade-off curve for satellite plant

25

461

r  0.01

20





0

z2 dt

15

10

5

0

r1

0

5

r  100 10



15

20

25

2

u dt

0

Figure 7.23 Nyquist plot for LQR design

1.5

Imaginary axis

1.0 0.5 0 0.5 1.0 1.5

2.5 2.0 1.5 1.0 0.5 Real axis

0

0.5

1.0

For the specified output of 2 × position + velocity, we let the tracking error be z=[ 2

1 ]x.

(7.122)

We then calculate from Eqs. (7.121) and (7.122) so that G0 (s) = −

s+2 . − ω02

s2

(7.123)

The symmetric 0◦ loci are shown in Fig. 7.24. The MATLAB statements to generate the SRL are (for ωo = 1), numGG=conv(−[1 2],−[−1 2]); denGG=conv([1 0 −1],[1 0 −1]); sysGG=tf(numGG,denGG); rlocus(sysGG);

462 Chapter 7 State-Space Design Figure 7.24 SRL for the inverted pendulum

Im(s) 3 2 1

4

3

2

2

3

4

Re(s)

1 2 3

Figure 7.25 Step response for the inverted pendulum

1.0 0.9 0.8 Position, x1

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

0.5

1

1.5

2 2.5 Time (sec)

3

3.5

4

4.5

For ρ = 1, we find that the closed-loop poles are at −1.36 ± j0.606, corresponding to K = [ −2.23 −2.73 ]. If we substitute the system matrices of this example into the equation for the input gains, Eq. (7.100), we find that the solution is   1 Nx = , 0 Nu = 1, N¯ = −1.23. With these values, the expression for the control using Nx and Nu (Eq. 7.101b) the controller reduces to ¯ u = −Kx + Nr. The corresponding step response for position is shown in Fig. 7.25.

7.6 Selection of Pole Locations for Good Design

463

As a final example in this section, we consider again the tape servomotor and introduce LQR design using the computer directly to solve for the optimal control law. From Eqs. (7.109) and (7.111), we know that the information needed to find the optimal control is given by the system matrices F and G and the output matrix H1 . Most computer-aided software packages, including MATLAB, use a more general form of Eq. (7.109):

∞ (xT Qx + uT Ru) dt. (7.124) J = 0

MATLAB lqr

Equation (7.124) reduces to the simpler form of Eq. (7.109) if we take Q = ρH1T H1 and R = 1. The direct solution for the optimal control gain is the MATLAB statement K = lqr(F, G, Q, R).

Bryson’s rule

(7.125)

One reasonable method to start the LQR design iteration is suggested by Bryson’s rule (Bryson and Ho, 1969). In practice, an appropriate choice to obtain acceptable values of x and u is to initially choose diagonal matrices Q and R such that Qii = 1/maximum acceptable value of [xi2 ], Rii = 1/maximum acceptable value of [ui2 ]. The weighting matrices are then modified during subsequent iterations to achieve an acceptable trade-off between performance and control effort.

EXAMPLE 7.24

LQR Design for a Tape Drive (a) Find the optimal control for the tape drive of Example 7.11, using the position x3 as the output for the performance index. Let ρ = 1. Compare the results with that of dominant second order obtained before. (b) Compare the LQR designs for ρ = 0.1, 1, 10. Solution (a) All we need to do here is to substitute the matrices into Eq. (7.125), form the feedback system, and plot the response. The performance index matrix is the scalar R = 1; the most difficult part of the problem is finding the statecost matrix Q. With the output-cost variable z = x3 , the output matrix from Example 7.11 is H3 = [ 0.5 0 0.5 0 0 ], and with ρ = 1, the required matrix is Q = H3T H3 ⎡ 0.25 ⎢ 0 ⎢ =⎣ 0.25 0

0 0 0 0

0.25 0 0.25 0

0 0 0 0

⎤ 0 0 ⎥ ⎥. 0 ⎦ 0

The gain is given by MATLAB, using the following statements:

464 Chapter 7 State-Space Design F=[0 2 0 0 0; −.1 −.35 .1 .1.75;0 0 0 2 0;.4 .4 −.4 −1.4;0 −.03 0 0 −1]; G=[0; 0; 0; 0; 1]; H3=[.5 0 .5 0 0]; R=1; rho=1; Q=rho*H3’*H3; K=lqr(F,G,Q,R) The MATLAB calculated gain is K = [ 0.6526

2.1667

0.3474

0.5976

1.0616 ].

(7.126)

The results of a position step and the corresponding tension are plotted in Figs. 7.18 and 7.19 (using step) with the earlier responses for comparison. Obviously, there is a vast range of choice for the elements of Q and R, so substantial experience is needed in order to use the LQR method effectively. (b) The LQR designs may be repeated as in part (a) with the same Q and R, but with ρ = 0.1, 10. Figure 7.26 shows a comparison of position step and the corresponding tension for the three designs. As seen from the results, the smaller values of ρ correspond to higher cost on the control and slower response, whereas the larger values of ρ correspond to lower cost on the control and relatively fast response.

Limiting Behavior of LQR Regulator Poles It is interesting to consider the limiting behavior of the optimal closed-loop poles as a function of the root-locus parameter (i.e., ρ) although, in practice, neither case would be used. “Expensive control” case (ρ → 0): Equation (7.109) primarily penalizes the use of control energy. If the control is expensive, then the optimal control does not move any of the open-loop poles except for those that are in the RHP. The poles in the RHP are simply moved to their mirror images in the LHP. The optimal control does this to stabilize the system using minimum control effort and makes no attempt to move any of the poles of the system in the LHP. The closed-loop pole locations are simply the starting points on the SRL in the LHP. The optimal control does not speed up the response of the system in this case. For the satellite plant, the vertical dashed line in Fig. 7.22 corresponds to the “expensive control” case and illustrates that the very low control usage results in a very large error in z. “Cheap control” case (ρ → ∞): In this case control energy is no object and arbitrary control effort may be used by the optimal control law. The control law then moves some of the closed-loop pole locations right on top of the zeros in the LHP. The rest are moved to infinity along the SRL asymptotes. If the system is nonminimum phase, some of the closed-loop poles are moved to mirror images of these zeros in the LHP, as shown in Example 7.23. The rest of the poles go to infinity and do so along a Butterworth filter pole pattern, as shown in Example 7.22. The optimal control law provides the fastest possible response time consistent with the LQR cost function. The feedback gain matrix K becomes unbounded in this case. For the double-integrator

7.6 Selection of Pole Locations for Good Design Figure 7.26 (a) Step response of the tape servomotor for LQR designs; (b) corresponding tension for tape servomotor step responses

465

1.4

Tape position, x3

1.2

r1

r  10

1.0 0.8 0.6 0.4

r  0.1

0.2 0

0

2

4

6 8 Time (msec)

10

12

10

12

(a) 0.10 0.05

r  10

Tape tension, T

0 0.05 0.10 r1

0.15 0.20 0.25

r  0.1 0

2

4

6 8 Time (msec) (b)

plant, the horizontal dashed line in Fig. 7.22 corresponds to the “cheap control” case.

Robustness Properties of LQR Regulators It has been proved (Anderson and Moore, 1990) that the Nyquist plot for LQR design avoids a circle of unity radius centered at the −1 point as shown in Fig. 7.23. This leads to extraordinary phase and gain margin properties. It can be shown (Problem 7.32) that the return difference must satisfy |1 + K(jωI − F)−1 G| ≥ 1.

(7.127)

Let us rewrite the loop gain as the sum of its real and imaginary parts: L(jω) = K(jωI − F)−1 G = Re(L(jω)) + jIm(L(jω)).

(7.128)

466 Chapter 7 State-Space Design Equation (7.127) implies that ([Re(L(jω)]+1)2 + [Im(L(jω)]2 ≥ 1,

LQR gain and phase margins

(7.129)

which means that the Nyquist plot must indeed avoid a circle centered at −1 with unit radius. This implies that 21 < GM < ∞, which means that the “upward” gain margin is GM = ∞ and the “downward” gain margin is GM = 21 (see also Problem 6.24 of Chapter 6). Hence the LQR gain matrix, K, can be multiplied by a large scalar or reduced by half with guaranteed closed-loop system stability. The phase margin, PM, is at least ±60◦ . These margins are remarkable, and it is not realistic to assume that they can be achieved in practice, because of the presence of modeling errors and lack of sensors!

7.6.3 Comments on the Methods The two methods of pole selection described in Sections 7.6.1 and 7.6.2 are alternatives the designer can use for an initial design by pole placement. Note that the first method (dominant second order) suggests selecting closed-loop poles without regard to the effect on the control effort required to achieve that response. In some cases, therefore, the resulting control effort may be ridiculously high. The second method (SRL), on the other hand, selects poles that result in some balance between system errors and control effort. The designer can easily examine the relationship between shifts in that balance (by changing ρ) and system root locations, time response, and feedback gains. Whatever initial pole-selection method we use, some modification is almost always necessary to achieve the desired balance of bandwidth, overshoot, sensitivity, control effort, and other practical design requirements. Further insight into pole selection will be gained from the examples that illustrate compensation in Section 7.8 and from the case studies in Chapter 10.

7.7 Estimator Design The control law designed in Section 7.5 assumed that all the state-variables are available for feedback. In most cases, not all the state-variables are measured. The cost of the required sensors may be prohibitive, or it may be physically impossible to measure all of the state-variables, as in, for example, a nuclear power plant. In this section we demonstrate how to reconstruct all of the state-variables of a system from a few measurements. If the estimate of the state is denoted by xˆ , it would be convenient if we could replace the true state in the control law given by Eq. (7.102) with the ¯ This is indeed possible, as we estimates, so that the control becomes u = −Kxˆ + Nr. shall see in Section 7.8, so construction of a state estimate is a key part of state-space control design.

7.7.1 Full-Order Estimators One method of estimating the state is to construct a full-order model of the plant dynamics, x˙ˆ = Fˆx + Gu, (7.130) where xˆ is the estimate of the actual state x. We know F, G, and u(t). Hence this estimator will be satisfactory if we can obtain the correct initial condition x(0) and set

7.7 Estimator Design Figure 7.27 Open-loop estimator

u

Process (F, G)

x

Model (F, G)

xˆ

H

y

H

yˆ

467

xˆ (0) equal to it. Figure 7.27 depicts this open-loop estimator. However, it is precisely the lack of information about x(0) that requires the construction of an estimator. Otherwise, the estimated state would track the true state exactly. Thus, if we made a poor estimate for the initial condition, the estimated state would have a continually growing error or an error that goes to zero too slowly to be of use. Furthermore, small errors in our knowledge of the system (F, G) would also cause the estimate to diverge from the true state. To study the dynamics of this estimator, we define the error in the estimate to be x˜  x − xˆ . Then the dynamics of this error system are given by x˜˙ = F˜x, x˜ (0) = x(0) − xˆ (0).

Feed back the output error to correct the state estimate equation.

(7.131)

(7.132)

The error converges to zero for a stable system (F stable), but we have no ability to influence the rate at which the state estimate converges to the true state. Furthermore, the error is converging to zero at the same rate as the natural dynamics of F. If this convergence rate were satisfactory, no control or estimation would be required. We now invoke the golden rule: When in trouble, use feedback. Consider feeding back the difference between the measured and estimated outputs and correcting the model continuously with this error signal. The equation for this scheme, shown in Fig. 7.28, is x˙ˆ = Fˆx + Gu + L(y − Hxˆ ). (7.133) Here L is a proportional gain defined as L = [l1 , l2 , . . . , ln ]T

(7.134)

and is chosen to achieve satisfactory error characteristics. The dynamics of the error can be obtained by subtracting the estimate [Eq. (7.133)] from the state [Eq. (7.44)], to get the error equation x˙˜ = (F − LH)˜x. (7.135) Estimate-error characteristic equation

The characteristic equation of the error is now given by det[sI − (F − LH)] = 0.

(7.136)

If we can choose L so that F − LH has stable and reasonably fast eigenvalues, then x˜ will decay to zero and remain there—independent of the known forcing function u(t) and its effect on the state x(t) and irrespective of the initial condition x˜ (0). This means that xˆ (t) will converge to x(t), regardless of the value of xˆ (0); furthermore, we can choose the dynamics of the error to be stable as well as much faster than the open-loop dynamics determined by F. Note that in obtaining Eq. (7.135), we have assumed that F, G, and H are identical in the physical plant and in the computer implementation of the estimator. If we do

468 Chapter 7 State-Space Design Figure 7.28 Closed-loop estimator

u(t)

Process (F, G)

x(t)

Model (F, G)

ˆx(t)

H

ˆy(t) 

H

y(t) 




L

not have an accurate model of the plant (F, G, H), the dynamics of the error are no longer governed by Eq. (7.135). However, we can typically choose L so that the error system is still at least stable and the error remains acceptably small, even with (small) modeling errors and disturbing inputs. It is important to emphasize that the nature of the plant and the estimator are quite different. The plant is a physical system such as a chemical process or servomechanism, whereas the estimator is usually a digital processor computing the estimated state according to Eq. (7.133). The selection of L can be approached in exactly the same fashion as K is selected in the control-law design. If we specify the desired location of the estimator error poles as si = β1 , β2 , . . . , βn , then the desired estimator characteristic equation is

αe (s) = (s − β1 )(s − β2 ) · · · (s − βn ).

(7.137)

We can then solve for L by comparing coefficients in Eqs. (7.136) and (7.137).

EXAMPLE 7.25

An Estimator Design for a Simple Pendulum Design an estimator for the simple pendulum. Compute the estimator gain matrix that will place both the estimator error poles at −10ω0 (five times as fast as the controller poles selected in Example 7.15). Verify the result using MATLAB for ω0 = 1. Evaluate the performance of the estimator. Solution. The equations of motion are     0 1 0 ˙x = x+ u, 1 −ω02 0 y = [1

0]x.

(7.138a) (7.138b)

We are asked to place the two estimator error poles at −10ω0 . The corresponding characteristic equation is αe (s) = (s + 10ω0 )2 = s2 + 20ω0 s + 100ω02 .

(7.139)

From Eq. (7.136), we get det[sI − (F − LH)] = s2 + l1 s + l2 + ω02 . Comparing the coefficients in Eqs. (7.139) and (7.140), we find that     20ω0 l1 L= = . l2 99ω02

(7.140)

(7.141)

7.7 Estimator Design

469

The result can also be found from MATLAB for ω0 = 1, using the following MATLAB statements: wo=1; F=[0 1;−wo*wo 0]; H=[1 0]; pe=[−10*wo;−10*wo]; Lt=acker(F’,H’,pe); L=Lt’ This yields L = [20 99]T and agrees with the preceding hand calculations. Performance of the estimator can be tested by adding the actual state feedback to the plant and plotting the estimate errors. Note that this is not the way the system will ultimately be built, but this approach provides a means of validating the estimator performance. Combining Eq. (7.71) of the plant with state feedback with Eq. (7.133) of the estimator with output feedback results in the following overall system equations:      x˙ F − GK 0 x = , (7.142) LH − GK F − LH xˆ x˙ˆ   x y=[ H 0 ] , (7.143) xˆ   x y˜ = [ H −H ] . (7.144) xˆ

MATLAB impulse, initial

A block diagram of the setup is drawn in Fig. 7.29. The response of this closed-loop system with ω0 = 1 to an initial condition x0 = [1.0, 0.0]T and xˆ 0 = [0, 0]T is shown in Fig. 7.30, where K is obtained from Example 7.15 and L comes from Eq. (7.141). The response may be obtained using impulse or initial in MATLAB. Note that the state estimates converge to the actual state-variables after an initial transient even though the initial value of xˆ had a large error. Also note that the estimate error decays approximately five times faster than the decay of the state itself, as we designed it to do.

Observer Canonical Form As was the case for control-law design, there is a canonical form for which the estimator gain design equations are particularly simple and the existence of a solution is obvious. We introduced this form in Section 7.4.1. The equations are in the observer Figure 7.29 Estimator connected to the plant

u

Plant r

N






u

x  Fx  Gu y y  Hx  x ⴚK




~ y

Estimator xˆ  Fxˆ  Gu  Ly~

xˆ

H

ˆy

470 Chapter 7 State-Space Design Figure 7.30 Initial-condition response of oscillator showing x and xˆ

3.5 3.0 2.5

ˆx2

Amplitude

2.0 1.5

xˆ1

1.0 x1

0.5 0

x2

0.5 1.0

Figure 7.31 Block diagram for observer canonical form of a third-order system

0

0.5

1.0

1.5 2.0 2.5 Time (sec)

3.0

3.5

4.0

U b3 




b2

1 s

x3o 

b1

 



x2o

1 s






a2

a3




1 s

x1o

Y

a1

canonical form and have the structure: x˙ o = Fo xo + Go u,

(7.145a)

y = Ho xo , where

Observer canonical form

⎡

−a1

⎢ ⎢ −a2 Fo = ⎢ ⎢ . ⎣ .. −an

1

0

0

0 .. .

1 .. .

0

Ho = [ 1

0

0

0

0

⎤ 0 . ⎥ . . . .. ⎥ ⎥, ⎥ 1 ⎦ 0

...

(7.145b) ⎡ ⎢ ⎢ Go = ⎢ ⎣

b1 b2 .. .

⎤ ⎥ ⎥ ⎥, ⎦

bn

. . . 0 ].

A block diagram for the third-order case is shown in Fig. 7.31. In observer canonical form, all the feedback loops come from the output, or observed signal. Like control canonical form, observer canonical form is a “direct” form because the values of the significant elements in the matrices are obtained directly from the coefficients of the numerator and denominator polynomials of the corresponding transfer function

7.7 Estimator Design

471

G(s). The matrix Fo is called a left companion matrix to the characteristic equation because the coefficients of the equation appear on the left side of the matrix. One of the advantages of the observer canonical form is that the estimator gains can be obtained from it by inspection. The estimator error closed-loop matrix for the third-order case is ⎡ ⎤ −a1 − l1 1 0 Fo − LHo = ⎣ −a2 − l2 0 1 ⎦ , (7.146) −a3 − l3 0 0 which has the characteristic equation s3 + (a1 + l1 )s2 + (a2 + l2 )s + (a3 + l3 ) = 0,

Observability

Observability matrix

Ackermann’s estimator formula

(7.147)

and the estimator gain can be found by comparing the coefficients of Eq. (7.147) with αe (s) from Eq. (7.137). In a development exactly parallel with the control-law case, we can find a transformation to take a given system to observer canonical form if and only if the system has a structural property that in this case we call observability. Roughly speaking, observability refers to our ability to deduce information about all the modes of the system by monitoring only the sensed outputs. Unobservability results when some mode or subsystem is disconnected physically from the output and therefore no longer appears in the measurements. For example, if only derivatives of certain state-variables are measured, and these state-variables do not affect the dynamics, a constant of integration is obscured. This situation occurs with a plant having the transfer function 1/s2 if only velocity is measured, for then it is impossible to deduce the initial value of the position. On the other hand, for an oscillator, a velocity measurement is sufficient to estimate position because the acceleration, and consequently the velocity observed, are affected by position. The mathematical test for determining observability is that the observability matrix ⎡ ⎤ H ⎢ HF ⎥ ⎢ ⎥ (7.148) O = ⎢ .. ⎥ ⎣ . ⎦ HFn−1 must have independent columns. In the one output case we will study, O is square, so the requirement is that O be nonsingular or have a nonzero determinant. In general, we can find a transformation to observer canonical form if and only if the observability matrix is nonsingular. Note that this is analogous to our earlier conclusions for transforming system matrices to control canonical form. As with control-law design, we could find the transformation to observer form, compute the gains from the equivalent of Eq. (7.147), and transform back. An alternative method of computing L is to use Ackermann’s formula in estimator form, which is ⎡ ⎤ 0 ⎢ 0 ⎥ ⎢ ⎥ (7.149) L = αe (F)O−1 ⎢ . ⎥ , ⎣ .. ⎦ 1 where O is the observability matrix given in Eq. (7.148).

472 Chapter 7 State-Space Design

TABLE 7.1

Duality Control

Estimation

F G H

FT HT GT

Duality

Duality of estimation and control

MATLAB acker, place

You may already have noticed from this discussion the considerable resemblance between estimation and control problems. In fact, the two problems are mathematically equivalent. This property is called duality. Table 7.1 shows the duality relationships between the estimation and control problems. For example, Ackermann’s control formula [Eq. (7.91)] becomes the estimator formula Eq. (7.149) if we make the substitutions given in Table 7.1. We can demonstrate this directly using matrix algebra. The control problem is to select the row matrix K for satisfactory placement of the poles of the system matrix F − GK; the estimator problem is to select the column matrix L for satisfactory placement of the poles of F − LH. However, the poles of F − LH equal those of (F − LH)T = FT − HT LT , and in this form, the algebra of the design for LT is identical to that for K. Therefore, where we used Ackermann’s formula or the place algorithm in the forms K=acker(F, G, pc ), K=place(F, G, pc ), for the control problem, we use Lt=acker(F
, H
, pe ), Lt=place(F
, H
, pe ), L=Lt
, where pe is a vector containing the desired estimator error poles for the estimator problem. Thus duality allows us to use the same design tools for estimator problems as for control problems with proper substitutions. The two canonical forms are also dual, as we can see by comparing the triples (Fc , Gc , Hc ) and (F◦ , G◦ , H◦ ).

7.7.2 Reduced-Order Estimators The estimator design method described in Section 7.7.1 reconstructs the entire state vector using measurements of some of the state-variables. If the sensors have no noise, then a full-order estimator contains redundancies, and it seems reasonable to question the necessity for estimating state-variables that are measured directly. Can we reduce the complexity of the estimator by using the state-variables that are measured directly and exactly? Yes. However, it is better to implement a full-order estimator if there is significant noise on the measurements because, in addition to estimating unmeasured state-variables, the estimator filters the measurements.

7.7 Estimator Design

473

The reduced-order estimator reduces the order of the estimator by the number (1 in this text) of sensed outputs. To derive this estimator, we start with the assumption that the output equals the first state as, for example, y = xa . If this is not so, a preliminary step is required. Transforming to observer form is possible but overkill; any nonsingular transformation with H as the first row will do. We now partition the state vector into two parts: xa , which is directly measured, and xb , which represents the remaining state-variables that need to be estimated. If we partition the system matrices accordingly, the complete description of the system is given by        Faa Fab xa Ga x˙ a = + u, (7.150a) x˙ b Fba Fbb xb Gb   xa y=[ 1 0 ] . (7.150b) xb The dynamics of the unmeasured state-variables are given by x˙ b = Fbb xb + Fba xa + Gb u, -. / ,

(7.151)

known input

where the right-most two terms are known and can be considered as an input into the xb dynamics. Because xa = y, the measured dynamics are given by the scalar equation x˙ a = y˙ = Faa y + Fab xb + Ga u. (7.152) If we collect the known terms of Eq. (7.152) on one side, yielding y˙ − Faa y − Ga u = Fab xb , -. / ,

(7.153)

known measurement

we obtain a relationship between known quantities on the left side, which we consider measurements, and unknown state-variables on the right. Therefore, Eqs. (7.152) and (7.153) have the same relationship to the state xb that the original equation [Eq. (7.150b)] had to the entire state x. Following this line of reasoning, we can establish the following substitutions in the original estimator equations to obtain a (reduced-order) estimator of xb : x ← xb ,

(7.154a)

F ← Fbb ,

(7.154b)

Gu ← Fba y + Gb u, y ← y˙ − Faa y − Ga u, H ← Fab .

(7.154c) (7.154d) (7.154e)

Therefore, the reduced-order estimator equations are obtained by substituting Eqs. (7.154) into the full-order estimator [Eq. (7.133)]: (7.155) x˙ˆ b = Fbb xˆ b + Fba y + Gb u + L (˙y − Faa y − Ga u − Fab xˆ b ). -. / -. / , , measurement

input

If we define the estimator error to be x˜ b  xb − xˆ b ,

(7.156)

474 Chapter 7 State-Space Design Figure 7.32 Reduced-order estimator structure

y Fba  LFaa  u

Gb  LGa






L

ˆxb  Ly xc

1 s

ˆxb  Ly xc








ˆxb

Fbb  LFab

then the dynamics of the error are given by subtracting Eq. (7.151) from Eq. (7.155) to get (7.157) x˙˜ b = (Fbb − LFab )˜xb , and its characteristic equation is given by det[sI − (Fbb − LFab )] = 0.

(7.158)

We design the dynamics of this estimator by selecting L so that Eq. (7.158) matches a reduced order αe (s). Now Eq. (7.155) can be rewritten as x˙ˆ b = (Fbb − LFab )ˆxb + (Fba − LFaa )y + (Gb − LGa )u + L˙y. (7.159) The fact that we must form the derivative of the measurements in Eq. (7.159) appears to present a practical difficulty. It is known that differentiation amplifies noise, so if y is noisy, the use of y˙ is unacceptable. To get around this difficulty, we define the new controller state to be (7.160) xc  xˆ b − Ly. In terms of this new state, the implementation of the reduced-order estimator is given by x˙ c = (Fbb − LFab )ˆxb + (Fba − LFaa )y + (Gb − LGa )u,

(7.161)

and y˙ no longer appears directly. A block-diagram representation of the reduced-order estimator is shown in Fig. 7.32.

EXAMPLE 7.26

A Reduced-Order Estimator Design for Pendulum

Reduced-order estimator

Design a reduced-order estimator for the pendulum that has the error pole at −10ω0 . Solution. We are given the system equations        0 1 x˙ 1 0 x1 = + u, x˙ 2 x2 1 −ω02 0   x1 y=[ 1 0 ] . x2 The partitioned matrices are  Faa Fba





0 1 = −ω02 0     0 Ga = . Gb 1 Fab Fbb

 ,

7.7 Estimator Design

475

From Eq. (7.158), we find the characteristic equation in terms of L: s − (0 − L) = 0. We compare it with the desired equation, αe (s) = s + 10ω0 = 0, which yields L = 10ω0 . The estimator equation, from Eq. (7.161), is x˙ c = −10ω0 xˆ 2 − ω02 y + u, and the state estimate, from Eq. (7.160), is xˆ 2 = xc + 10ω0 y.

MATLAB impulse, initial

We use the control law given in the earlier examples. The response of the estimator to a plant initial condition x0 = [1.0, 0.0]T and an estimator initial condition xc0 = 0 is shown in Fig. 7.33 for ω0 = 1. The response may be obtained using impulse or initial in MATLAB. Note the similarity of the initial-condition response to that of the full-order estimator plotted in Fig. 7.30. The reduced-order estimator gain can also be found from MATLAB by using











Lt = acker(F bb , F ab , pe ), Lt = place(F bb , F ab , pe ), L = Lt
. The conditions for the existence of the reduced-order estimator are the same as for the full-order estimator—namely, observability of (F, H). Figure 7.33 Initial-condition response of the reduced-order estimator

10 8

Amplitude

6 4

xˆ2

2 x1 0 x2 2

0

0.5

1.0

1.5

2.0 2.5 Time (sec)

3.0

3.5

4.0

476 Chapter 7 State-Space Design 7.7.3 Estimator Pole Selection Design rules of thumb for selecting estimator poles

Process noise

We can base our selection of estimator pole locations on the techniques discussed in Section 7.6 for the case of controller poles. As a rule of thumb, the estimator poles can be chosen to be faster than the controller poles by a factor of 2 to 6. This ensures a faster decay of the estimator errors compared with the desired dynamics, thus causing the controller poles to dominate the total response. If sensor noise is large enough to be a major concern, we may choose the estimator poles to be slower than two times the controller poles, which would yield a system with lower bandwidth and more noise smoothing. However, we would expect the total system response in this case to be strongly influenced by the location of the estimator poles. If the estimator poles are slower than the controller poles, we would expect the system response to disturbances to be dominated by the dynamic characteristics of the estimator rather than by those selected by the control law. In comparison with the selection of controller poles, estimator pole selection requires us to be concerned with a much different relationship than with control effort. As in the controller, there is a feedback term in the estimator that grows in magnitude as the requested speed of response increases. However, this feedback is in the form of an electronic signal or a digital word in a computer, so its growth causes no special difficulty. In the controller, increasing the speed of response increases the control effort; this implies the use of a larger actuator, which in turn increases size, weight, and cost. The important consequence of increasing the speed of response of an estimator is that the bandwidth of the estimator becomes higher, thus causing more sensor noise to pass on to the control actuator. Of course, if (F, H) are not observable, then no amount of estimator gain can produce a reasonable state estimate. Thus, as with controller design, the best estimator design is a balance between good transient response and low-enough bandwidth that sensor noise does not significantly impair actuator activity. Both dominant second-order and prototype characteristic equation ideas can be used to meet the requirements. There is also a result for estimator gain design based on the SRL. In optimal estimation theory, the best choice for estimator gain is dependent on the ratio of sensor noise intensity ν to process (disturbance) noise intensity [w in Eq. (7.163)]. This is best understood by reexamining the estimator equation xˆ˙ = Fˆx + Gu + L(y − Hxˆ ) (7.162) to see how it interacts with the system when process noise w is present. The plant with process noise is described by x˙ = Fx + Gu + G1 w,

Sensor noise

(7.163)

and the measurement equation with sensor noise ν is described by y = Hx + ν.

(7.164)

The estimator error equation with these additional inputs is found directly by subtracting Eq. (7.162) from Eq. (7.163) and substituting Eq. (7.164) for y: (7.165) x˙˜ = (F − LH)˜x + G1 w − Lν. In Eq. (7.165) the sensor noise is multiplied by L and the process noise is not. If L is very small, then the effect of sensor noise is removed, but the estimator’s dynamic

7.7 Estimator Design

Estimator SRL equation

477

response will be “slow,” so the error will not reject effects of w very well. The state of a low-gain estimator will not track uncertain plant inputs very well. These results can, with some success, also be applied to model errors in, for example, F or G. Such model errors will add terms to Eq. (7.165) and act like additional process noise. On the other hand, if L is large, then the estimator response will be fast and the disturbance or process noise will be rejected, but the sensor noise, multiplied by L, results in large errors. Clearly, a balance between these two effects is required. It turns out that the optimal solution to this balance can be found under very reasonable assumptions by solving an SRL equation for the estimator that is very similar to the one for the optimal control formulation [Eq. (7.112)]. The estimator SRL equation is 1 + qGe (−s)Ge (s) = 0,

(7.166)

where q is the ratio of input disturbance noise intensity to sensor noise intensity and Ge is the transfer function from the process noise to the sensor output and is given by Ge (s) = H(sI − F)−1 G1 .

(7.167)

Note from Eqs. (7.112) and (7.166) that Ge (s) is similar to G0 (s). However, a comparison of Eqs. (7.113) and (7.167) shows that Ge (s) has the input matrix G1 instead of G, and that G0 is the transfer function from the control input u to cost output z and has output matrix H1 instead of H. The use of the estimator SRL [Eq. (7.166)] is identical to the use of the controller SRL.A root locus with respect to q is generated, thus yielding sets of optimal estimator poles corresponding more or less to the ratio of process noise intensity to sensor noise intensity. The designer then picks the set of (stable) poles that seems best, considering all aspects of the problem. An important advantage of using the SRL technique is that after the process noise input matrix G1 has been selected, the arbitrariness is reduced to one degree of freedom, the selection q, instead of the many degrees of freedom required to select the poles directly in a higher-order system. A final comment concerns the reduced-order estimator. Because of the presence of a direct transmission term from y through L to xb (see Fig. 7.32), the reduced-order estimator has a much higher bandwidth from sensor to control when compared with the full-order estimator. Therefore, if sensor noise is a significant factor, the reducedorder estimator is less attractive because the potential saving in complexity is more than offset by the increased sensitivity to noise.

EXAMPLE 7.27

SRL Estimator Design for a Simple Pendulum Draw the estimator SRL for the linearized equations of the simple inverted pendulum with ωo = 1. Take the output to be a noisy measurement of position with noise intensity ratio q. Solution. We are given the system equations        0 1 x˙ 1 x1 0 = + w, x˙ 2 x2 1 −ω02 0   x1 + v. y=[ 1 0 ] x2

Figure 7.34 Symmetric root locus for the inverted pendulum estimator design

Imaginary axis

478 Chapter 7 State-Space Design 5 4 3 2 1 0 1 2 3 4 5

q0

q0

5 4 3 2 1 0 1 Real axis

2

3

4

5

We then calculate from Eq. (7.167) that 1 . + ω02 The symmetric 180◦ loci are shown in Fig. 7.34. The MATLAB statements to generate the SRL are (for ωo = 1) Ge (s) =

s2

numGG=1; denGG=conv([1 0 1],[1 0 1]); sysGG=tf(numGG,denGG); rlocus(sysGG); We would choose two stable roots for a given value of q, for example s = −3±j3.18 for q = 365, and use them for estimator pole placement.

7.8 Compensator Design: Combined Control Law and Estimator Regulator

If we take the control-law design described in Section 7.5, combine it with the estimator design described in Section 7.7, and implement the control law by using the estimated state-variables, the design is complete for a regulator that is able to reject disturbances but has no reference input to be tracked. However, because the control law was designed for feedback of the actual (not the estimated) state, you may wonder what effect using xˆ in place of x has on the system dynamics. In this section we compute this effect. In doing so we will compute the closed-loop characteristic equation and the open-loop compensator transfer function. We will use these results to compare the state-space designs with root-locus and frequency-response designs. The plant equation with feedback is now x˙ = Fx − GKxˆ ,

(7.168)

which can be rewritten in terms of the state error x˜ as x˙ = Fx − GK(x − x˜ ).

(7.169)

7.8 Compensator Design: Combined Control Law and Estimator

479

The overall system dynamics in state form are obtained by combining Eq. (7.169) with the estimator error (Eq. 7.135) to get      x˙ F − GK GK x = . (7.170) x˜ 0 F − LH x˙˜ The characteristic equation of this closed-loop system is   sI − F + GK −GK det = 0. 0 sI − F + LH

(7.171)

Because the matrix is block triangular (see Appendix WE), we can rewrite Eq. (7.171) as det(sI − F + GK) · det(sI − F + LH) = αc (s)αe (s) = 0. Poles of the combined control law and estimator

(7.172)

In other words, the set of poles of the combined system consists of the union of the control poles and the estimator poles. This means that the designs of the control law and the estimator can be carried out independently, yet when they are used together in this way, the poles remain unchanged.9 To compare the state-variable method of design with the transform methods discussed in Chapters 5 and 6, we note from Fig. 7.35 that the blue shaded portion corresponds to a compensator. The state equation for this compensator is obtained by including the feedback law u = −Kxˆ (because it is part of the compensator) in the estimator Eq. (7.133) to get xˆ˙ = (F − GK − LH)ˆx + Ly, (7.173a) u = −Kxˆ .

(7.173b)

Note that Eq. (7.173a) has the same structure as Eq. (7.21a), which we repeat here: x˙ = Fx + Gu.

(7.174)

Because the characteristic equation of Eq. (7.21a) is det(sI − F) = 0, Figure 7.35 Estimator and controller mechanization

y

w

Sensor

Plant u(t)

(7.175)

x  Fx  Gu

x(t)

H

y(t)

u(t) Control law K

Estimator

ˆx(t) xˆ  Fxˆ  Gu ˆ  L(y  Hx) Compensator

9 This is a special case of the separation principle (Gunckel and Franklin, 1963), which holds in much more general contexts and allows us to obtain an overall optimal design by combining the separate designs of control law and estimator in certain stochastic cases.

480 Chapter 7 State-Space Design the characteristic equation of the compensator is found by comparing Eqs. (7.173a) and (7.174) and substituting the equivalent matrices into Eq. (7.175) to get det(sI − F + GK + LH) = 0.

Compensator transfer function

(7.176)

Note that we never specified the roots of Eq. (7.176) nor used them in our discussion of the state-space design technique. (Note also that the compensator is not guaranteed to be stable; the roots of Eq. (7.176) can be in the RHP.) The transfer function from y to u representing the dynamic compensator is obtained by inspecting Eq. (7.48) and substituting in the corresponding matrices from Eq. (7.173): Dc (s) =

U(s) = −K(sI − F + GK + LH)−1 L. Y (s)

(7.177)

The same development can be carried out for the reduced-order estimator. Here the control law is   xa (7.178) = −Ka y − Kb xˆ b . u = −[ Ka Kb ] xˆ b Substituting Eq. (7.178) into Eq. (7.174) and using Eq. (7.161) and some algebra, we obtain x˙ c = Ar xc + Br y,

(7.179a)

u = Cr xc + Dr y,

(7.179b)

where

Reduced-order compensator transfer function

Ar = Fbb − LFab − (Gb − LGa )Kb ,

(7.180a)

Br = Ar L + Fba − LFaa − (Gb − LGa )Ka ,

(7.180b)

Cr = −Kb ,

(7.180c)

Dr = −Ka − Kb L.

(7.180d)

The dynamic compensator now has the transfer function Dcr (s) =

U(s) = Cr (sI − Ar )−1 Br + Dr . Y (s)

(7.181)

When we compute Dc (s) or Dcr (s) for a specific case, we will find that they are very similar to the classical compensators given in Chapters 5 and 6, in spite of the fact that they are arrived at by entirely different means.

EXAMPLE 7.28

Full-Order Compensator Design for Satellite Attitude Control Design a compensator using pole placement for the satellite plant with transfer function 1/s2 . Place the control poles at s = −0.707 ± 0.707j(ωn = 1 rad/sec, ζ = 0.707) and place the estimator poles at ωn = 5 rad/sec, ζ = 0.5. Solution. A state-variable description for the given transfer function G(s) = 1/s2 is     0 1 0 x˙ = x+ u, 0 0 1  y = 1 0 x.

7.8 Compensator Design: Combined Control Law and Estimator

481

If we place the control roots at s = −0.707 ± 0.707j (ωn = 1 rad/sec, ζ = 0.7), then √ αc (s) = s2 + s 2 + 1. (7.182) From K = place(F,G,pc), the state feedback gain is found to be  √ K= 1 2 . If the estimator error roots are at ωn = 5 rad/sec and ζ = 0.5, then the desired estimator characteristic polynomial is αe (s) = s2 + 5s + 25 = s + 2.5 ± 4.3j,

(7.183)

and, from Lt = place(F’,H’,pe), the estimator feedback-gain matrix is found to be   5 L= . 25 The compensator transfer function given by Eq. (7.177) is Dc (s) = −40.4

(s + 0.619) , s + 3.21 ± 4.77j

(7.184)

which looks very much like a lead compensator in that it has a zero on the real axis to the right of its poles; however, rather than one real pole, Eq. (7.184) has two complex poles. The zero provides the derivative feedback with phase lead, and the two poles provide some smoothing of sensor noise. The effect of the compensation on this system’s closed-loop poles can be evaluated in exactly the same way we evaluated compensation in Chapters 5 and 6 using root-locus or frequency-response tools. The gain of 40.4 in Eq. (7.184) is a result of the pole selection inherent in Eqs. (7.182) and (7.183). If we replace this specific value of compensator gain with a variable gain K, then the characteristic equation for the closed-loop system of plant plus compensator becomes 1+K

Identical results of state-space and frequency response design methods

(s + 0.619) = 0. (s + 3.21 ± 4.77j)s2

(7.185)

The root-locus technique allows us to evaluate the roots of this equation with respect to K, as drawn in Fig. 7.36. Note that the locus goes through the roots selected for Eqs. (7.182) and (7.183), and, when K = 40.4, the four roots of the closed-loop system are equal to those specified. The frequency-response plots given in Fig. 7.37 show that the compensation designed using state-space accomplishes the same results that one would strive for using frequency-response design. Specifically, the uncompensated phase margin of 0◦ increases to 53◦ in the compensated case, and the gain K = 40.4 produces a crossover frequency ωc = 1.35 rad/sec. Both these values are roughly consistent with the controller closed-loop roots, with ωn = 1 rad/sec and ζ = 0.7, as we would expect, because these slow controller poles are dominant in the system response over the fast estimator poles.

482 Chapter 7 State-Space Design Figure 7.36 Root locus for the combined control and estimator,with process gain as the parameter

Im(s)

6 4

K  40.4

2 K  40.4 8

6

4

2

Re(s) 2 4

Magnitude

Figure 7.37 Frequency response for G(s) = 1/s2

100

40

10

20

db

6

0

1

20

0.1 0.01

0.1

1 v (rad/sec)

10

100

120

Phase

150

Compensated Uncompensated

53

180 210 240 270

0.02 0.04

0.1

0.2

0.4 0.6 1

2

4

6 10

20

40 60 100

v (rad/sec)

Now we consider a reduced-order estimator for the same system.

EXAMPLE 7.29

Reduced-Order Compensator Design for a Satellite Attitude Control Repeat the design for the 1/s2 satellite plant, but use a reduced-order estimator. Place the one estimator pole at −5 rad/sec. Solution. From Eq. (7.158) we know that the estimator gain is L = 5,

7.8 Compensator Design: Combined Control Law and Estimator Figure 7.38 Simplified block diagram of a reduced-order controller that is a lead network

Y

483

8.07 0.619 



1 s









U

6.41

Figure 7.39 Root locus of a reduced-order controller and 1/s2 process, root locations at K = 8.07 shown by the dots

Im(s) 4 2

8

6

4

2

Re(s)

and from Eqs. (7.179a, b) the scalar compensator equations are x˙ c = −6.41xc − 33.1y, u = −1.41xc − 8.07y, where, from Eq. (7.160), xc = xˆ 2 − 5y. The compensator has the transfer function calculated from Eq. (7.181) to be 8.07(s + 0.619) Dcr (s) = − s + 6.41 and is shown in Fig. 7.38. The reduced-order compensator here is precisely a lead network. This is a pleasant discovery, as it shows that transform and state-variable techniques can result in exactly the same type of compensation. The root locus of Fig. 7.39 shows that the closedloop poles occur at the assigned locations. The frequency response of the compensated system seen in Fig. 7.40 shows a phase margin of about 55◦ . As with the full-order estimator, analysis by other methods confirms the selected root locations. More subtle properties of the pole-placement method can be illustrated by a third-order system.

EXAMPLE 7.30

Full-Order Compensator Design for DC Servo Use the state-space pole-placement method to design a compensator for the DC servo system with the transfer function 10 . G(s) = s(s + 2)(s + 8)

Magnitude

100

40

10

20

db

484 Chapter 7 State-Space Design

0

1

20

0.1

0.01

0.1

1 v (rad/sec)

10

100

Phase

120 150

Compensated Uncompensated

55

180 210

0.02 0.04

0.1

0.2

0.4 0.6 1 2 v (rad/sec)

4

6 8 10

20

40 60 100

Figure 7.40 Frequency response for G(s) = 1/s2 with a reduced-order estimator Using a state description in observer canonical form, place the control poles at pc = [−1.42; −1.04 ± 2.14j] locations and the full-order estimator poles at pe = [−4.25; −3.13 ± 6.41j]. Solution. A block diagram of this system in observer Fig. 7.41. The corresponding state-space matrices are ⎡ ⎤ ⎡ −10 1 0 F = ⎣ −16 0 1 ⎦ , G = ⎣ 0 0 0  H= 1 0 0 , J = 0.

canonical form is shown in ⎤ 0 0 ⎦, 10

The desired poles are

 pc = −1.42; −1.04 + 2.14 ∗ j; −1.04 − 2.14 ∗ j .

We compute the state feedback gain to be K=(F,G,pc), K = [ −46.4

5.76

−0.65 ].

The estimator error poles are at pe = [−4.25; −3.13 + 6.41 ∗ j; −3.13 − 6.41 ∗ j]; We compute the estimator gain to be Lt=place(F’,H’,pe), L=Lt’, ⎡ ⎤ 0.5 L = ⎣ 61.4 ⎦ . 216

7.8 Compensator Design: Combined Control Law and Estimator Figure 7.41 DC servo in observer canonical form

U

1 s

10








1 s






1 s

Y

10

16

Figure 7.42 Root locus for DC servo pole assignment



485

Im(s) 8 6 4

8

6

4

2

2

Re(s)

4 6 8

The compensator transfer function, as given by substituting into Eq. (7.177), is (s + 0.432)(s + 2.10) . Dc (s) = −190 (s − 1.88)(s + 2.94 ± 8.32j) Figure 7.42 shows the root locus of the system of compensator and plant in series, plotted with the compensator gain as the parameter. It verifies that the roots are in the desired locations specified when the gain K = 190 in spite of the peculiar (unstable) compensation that has resulted. Even though this compensator has an unstable root at s = +1.88, all system closed-loop poles (controller and estimator) are stable.

Conditionally stable compensator

An unstable compensator is typically not acceptable because of the difficulty in testing either the compensator by itself or the system in open loop during a bench checkout. In some cases, however, better control can be achieved with an unstable compensator; then its inconvenience in checkout may be worthwhile.10 Figure 7.33 shows that a direct consequence of the unstable compensator is that the system becomes unstable as the gain is reduced from its nominal value. Such a system is called conditionally stable and should be avoided if possible. As we shall see in Chapter 9, actuator saturation in response to large signals has the effect of lowering the effective gain, and in a conditionally stable system, instability can result. 10 There are even systems that cannot be stabilized with a stable compensator.

486 Chapter 7 State-Space Design Also, if the electronics are such that the control amplifier gain rises continuously from zero to the nominal value during startup, such a system would be initially unstable. These considerations lead us to consider alternative designs for this system.

EXAMPLE 7.31

Redesign of the DC Servo System with a Reduced-Order Estimator Design a compensator for the DC servo system of Example 7.30 by using the same control poles but with a reduced-order estimator. Place the estimator poles at −4.24 ± 4.24j positions with ωn = 6 and ζ = 0.707. Solution. The reduced-order estimator corresponds to  pe = −4.24 + 4.24 ∗ j; −4.24 − 4.24 ∗ j After partitioning we have 

Faa Fba

Fab Fbb



⎡

−10 = ⎣ −16 0

1 0 0

⎤ 0 1 ⎦, 0

⎡

⎤ 0 ⎣ ⎦ = ⎣ 0 ⎦. 10 Gb Ga

⎤

⎡

Solving for the estimator error characteristic polynomial, det(sI − Fbb + LFab ) = αe (s), we find (using place) that

 L=

8.5 36

 .

The compensator transfer function, given by Eq. (7.181), is computed to be Dcr (s) = 20.93 A nonminimum-phase compensator

(s − 0.735)(s + 1.871) . (s + 0.990 ± 6.120j)

The associated root locus for this system is shown in Fig. 7.43. Note that this time we have a stable but nonminimum-phase compensator and a zero-degree root locus.

Figure 7.43 Root locus for DC servo reduced-order controller

Im(s) 6 5 4 3 1 8 7 6 5 4

2 1 1 3 4 5

1 2

3

Re(s)

7.8 Compensator Design: Combined Control Law and Estimator

487

The RHP portion of the locus will not cause difficulties because the gain has to be selected to keep all closed-loop poles in LHP. As a next pass at the design for this system, we attempt a design with the SRL.

EXAMPLE 7.32

Redesign of the DC Servo Compensator Using the SRL Design a compensator for the DC servo system of Example 7.30 using pole placement based on the SRL. For the control law, let the cost output z be the same as the plant output; for the estimator design, assume that the process noise enters at the same place as the system control signal. Select roots for a control bandwidth of about 2.5 rad/sec, and choose the estimator roots for a bandwidth of about 2.5 times faster than the control bandwidth (6.3 rad/sec). Derive an equivalent discrete controller with a sampling period of Ts = 0.1 sec (10 times the fastest pole), and compare the continuous and digital control outputs and control efforts. Solution. Because the problem has specified that G1 = G and H1 = H, then the SRL is the same for the control as for the estimator, so we need to generate only one locus based on the plant transfer function. The SRL for the system is shown in Fig. 7.44. From the locus, we select −2 ± 1.56j and −8.04 as the desired control poles (pc=[−2+1.56*j;−2−1.56*j;−8.04]) and −4±4.9j and −9.169 (pe=[−4+4.9*j;−4−4.9*j;−9.169]) as the desired estimator poles. The state feedback gain is K=(F,G,pc), or K = [ −0.285 0.219 0.204 ], and the estimator gain is Lt=place(F’,H’,pe), L=Lt’, or ⎤ ⎡ 7.17 L = ⎣ 97.4 ⎦ . 367 Notice that the feedback gains are much smaller than before. The resulting compensator transfer function is computed from Eq. (7.177) to be 94.5(s + 7.98)(s + 2.52) Dc (s) = − . (s + 4.28 ± 6.42j)(s + 10.6)

Figure 7.44 Symmetric root locus

Im(s) 5 4 3 2 1 9 8 7 6 5 4 3 2 1 1 2 Controller poles Estimator poles 3 4 5

1

2

3

4

5

6

7

8

Re(s)

488 Chapter 7 State-Space Design Figure 7.45 Root locus for pole assignment from the SRL

Im(s) 7 6 4 3 2 1 10

6

4

Controller poles Estimator poles

2

1 2 3 4

1

2

Re(s)

6 7

We now take this compensator, put it in series with the plant, and use the compensator gain as the parameter. The resulting ordinary root locus of the closed-loop system is shown in Fig. 7.45. When the root-locus gain equals the nominal gain of 94.5, the roots are at the closed-loop locations selected from the SRL, as they should be. Note that the compensator is now stable and minimum phase. This improved design comes about in large part because the plant pole at s = −8 is virtually unchanged by either controller or estimator. It does not need to be changed for good performance; in fact, the only feature in need of repair in the original G(s) is the pole at s = 0. Using the SRL technique, we essentially discovered that the best use of control effort is to shift the two low-frequency poles at s = 0 and −2 and to leave the pole at s = −8 virtually unchanged. As a result, the control gains are much lower and the compensator design is less radical. This example illustrates why LQR design is typically preferable over pole placement. The discrete equivalent for the controller is obtained from MATLAB with the c2d command, as in the following code: nc=94.5*conv([1 7.98],[1 2.52]); % form controller numerator dc=conv([1 8.56 59.5348],[1 10.6]); % form controller denominator sysDc=tf(nc,dc); % form controller system description ts=0.1;% sampling time of 0.1 sec sysDd=c2d(sysDc,ts,’zoh’); % convert controller to discrete time Discrete controller

The resulting controller has the discrete transfer function Dd (z) =

5.9157(z + 0.766)(z + 0.4586) . (z − 0.522 ± 0.3903j)(z + 0.3465)

The equation for the control law (with the sample period suppressed for clarity) is u(k + 1) = 1.3905u(k) − 0.7866u(k − 1) + 0.1472u(k − 2) + e(k) − 7.2445e(k − 2) + 2.0782e(k − 2).

7.8 Compensator Design: Combined Control Law and Estimator

  Step

Continuous controller

Plant

94.5s2  992.25s  1900.3572 3 s  19.16s2  150.27s  631.06

10 s3  10s2  16s

Mux

 

Control

Discrete controller

Plant 1

5.9157z2  7.2445z  2.0782 z3  1.3905z2  0.7866z  0.1472

10 s3  10s2  16s

Mux1

489

Output

Figure 7.46 SIMULINK
block diagram to compare continuous and discrete controllers

SIMULINK simulation

A SIMULINK
diagram for simulating both the continuous and discrete systems is shown in Fig. 7.46. A comparison of the continuous and discrete step responses and control signals is shown in Fig. 7.47. Better agreement between the two responses can be obtained if the sampling period is reduced. Armed with the knowledge gained from Example 7.32, let us go back, with a better selection of poles, to investigate the use of pole placement for this example. Initially we used the third-order locations, which produced three poles with a natural frequency of about 2 rad/sec. This design moved the pole at s = −8 to s = −1.4, thus violating the principle that open-loop poles should not be moved unless they are a problem. Now let us try it again, this time using dominant second-order locations to shift the slow poles, and leaving the fast pole alone at s = −8.

EXAMPLE 7.33

DC Servo System Redesign with Modified Dominant Second-Order Pole Locations Design a compensator for the DC servo system of Example 7.30 by using pole placement with control poles given by pc =[−1.41 ± 1.41 j; −8] and the estimator poles given by pe =[−4.24 ± 4.24j; −8] Solution. With these pole locations, we find that the required feedback gain is (using K=place(F,G,pc)) K = [ −0.469 0.234 0.0828 ], which has a smaller magnitude than the case where the pole at s = −8 was moved.

490 Chapter 7 State-Space Design 1.4 Digital controller 1.2 1.0 Continuous controller y

0.8 0.6 0.4 0.2 0

0

1

2 3 Time (sec)

4

5

4

5

(a) 7 6 5 4 3 u

Figure 7.47 Comparison of step responses and control signals for continuous and discrete controllers: (a) step responses; (b) control signals

2 Continuous controller 1 0 1 2

Digital controller 0

1

2 3 Time (sec) (b)

We find the estimator gain to be (using Lt = place(F’,H’,pe), L=Lt’) ⎡ ⎤ 6.48 L = ⎣ 87.8 ⎦ . 288 The compensator transfer function is 414(s + 2.78)(s + 8) , Dc (s) = − (s + 4.13 ± 5.29j)(s + 9.05) which is stable and minimum phase. This example illustrates the value of judicious pole selection and of the SRL technique. The poor pole selection inherent in the initial use of the poles results in higher control effort and produces an unstable compensator. Both of these undesirable features are eliminated by using the SRL (or LQR), or by improved pole selection. But

7.9 Introduction of the Reference Input with the Estimator

491

we really need to use SRL to guide the proper selection of poles. The bottom line is that SRL (or LQR) is the method of choice! As seen from some of the preceding examples, we have shown the use of optimal design via the SRL. However, it is more common in practice to skip that step and use LQR directly.

7.9 Introduction of the Reference Input with the Estimator The controller obtained by combining the control law studied in Section 7.5 with the estimator discussed in Section 7.8 is essentially a regulator design. This means that the characteristic equations of the control and the estimator are chosen for good disturbance rejection—that is, to give satisfactory transients to disturbances such as w(t). However, this design approach does not consider a reference input, nor does it provide for command following, which is evidenced by a good transient response of the combined system to command changes. In general, good disturbance rejection and good command following both need to be taken into account in designing a control system. Good command following is done by properly introducing the reference input into the system equations. Let us repeat the plant and controller equations for the full-order estimator; the reduced-order case is the same in concept, differing only in detail: Plant: x˙ = Fx + Gu, y = Hx; Controller: x˙ˆ = (F − GK − LH)ˆx + Ly, u = −Kxˆ .

(7.186a) (7.186b) (7.187a) (7.187b)

Figure 7.48 shows two possibilities for introducing the command input r into the system. This figure illustrates the general issue of whether the compensation should be put in the feedback or feed-forward path. The response of the system to command inputs is different, depending on the configuration, because the zeros of the transfer functions are different. The closed-loop poles are identical, however, as can be easily verified by letting r = 0 and noting that the systems are then identical. The difference in the responses of the two configurations can be seen quite easily. Consider the effect of a step input in r. In Fig. 7.48(a) the step will excite the estimator in precisely the same way that it excites the plant; thus the estimator error will remain zero during and after the step. This means that the estimator dynamics are not excited by the command input, so the transfer function from r to y must have zeros at the estimator pole locations that cancel those poles. As a result, a step command will excite system behavior that is consistent with the control poles alone—that is, with the roots of det(sI − F + GK) = 0. In Fig. 7.48(b), a step command in r enters directly only into the estimator, thus causing an estimation error that decays with the estimator dynamic characteristics in addition to the response corresponding to the control poles. Therefore, a step command will excite system behavior consistent with both control roots and estimator

492 Chapter 7 State-Space Design Figure 7.48 Possible locations for introducing the command input: (a) compensation in the feedback path; (b) compensation in the feed-forward path

Plant r

N



u




y

 Estimator

ⴚK

ˆx Compensator (a)

Estimator r






Plant xˆ

e

K

u

y

 Compensator

(b)

roots—that is, the roots of det(sI − F + GK) · det(sI − F + LH) = 0. For this reason, the configuration shown in Fig. 7.48(a) is typically the superior way to command the system, where N¯ is found using Eqs. (7.100)–(7.102). In Section 7.9.1, we will show a general structure for introducing the reference input with three choices of parameters that implement either the feed-forward or the feedback case. We will analyze the three choices from the point of view of the system zeros and the implications the zeros have for the system transient response. Finally, in Section 7.9.2 we will show how to select the remaining parameter to eliminate constant errors.

7.9.1 A General Structure for the Reference Input

Controller equations

Given a reference input r(t), the most general linear way to introduce r into the system equations is to add terms proportional to it in the controller equations. We can do this ¯ to Eq. (7.187b) and Mr to Eq. (7.187a). Note that in this case, N¯ is a by adding Nr scalar and M is an n×1 vector. With these additions, the controller equations become x˙ˆ = (F − GK − LH)ˆx + Ly + Mr,

(7.188a)

¯ u = −Kxˆ + Nr.

(7.188b)

The block diagram is shown in Fig. 7.49(a). The alternatives shown in Fig. 7.48 cor¯ Because r(t) is an external signal, it is clear respond to different choices of M and N. that neither M nor N¯ affects the characteristic equation of the combined controller– estimator system. In transfer-function terms, the selection of M and N¯ will affect

7.9 Introduction of the Reference Input with the Estimator Plant

Plant

u r

N



u

y




r



N

y




M

Estimator ⴚK

493

Estimator

xˆ

ⴚK

(a)

xˆ

(b)

Plant u

y 


Estimator ⴚK



r

e

xˆ

(c)

Figure 7.49 Alternative ways to introduce the reference input: (a) general case—zero assignment; (b) standard case—estimator not excited, zeros = αe (s); (c) error-control case—classical compensation

Three methods for selecting M and N

only the zeros of transmission from r to y and, as a consequence, can significantly affect the transient response but not the stability. How can we choose M and N¯ to obtain satisfactory transient response? We should point out that we assigned the poles of the system by feedback gains K and L, and we are now going to assign zeros by ¯ feed-forward gains M and N. ¯ There are three strategies for choosing M and N: 1. Autonomous estimator: Select M and N¯ so that the state estimator error equation is independent of r [Fig. 7.49(b)]. 2. Tracking-error estimator: Select M and N¯ so that only the tracking error, e = (r − y), is used in the control [Fig. 7.49(c)]. 3. Zero-assignment estimator: Select M and N¯ so that n of the zeros of the overall transfer function are assigned at places of the designer’s choice [Fig. 7.49(a)]. CASE 1. From the viewpoint of estimator performance, the first method is quite attractive and the most widely used of the alternatives. If xˆ is to generate a good estimate of x, then surely x˜ should be as free of external excitation as possible; that is, x˜ should be uncontrollable from r. The computation of M and N¯ to bring this about is quite easy. The estimator error equation is found by subtracting Eq. (7.188a)

494 Chapter 7 State-Space Design from Eq. (7.186a), with the plant output [Eq. (7.186b)] substituted into the estimator [Eq. (7.187a)] and the control [Eq. (7.187b)] substituted into the plant [Eq. (7.186a)]: ¯ − [(F − GK − LH)ˆx + Ly + Mr], (7.189a) x˙ − xˆ˙ = Fx + G(−Kxˆ + Nr) ¯ − Mr. x˙˜ = (F − LH)˜x + GNr

(7.189b)

If r is not to appear in Eq. (7.189a), then we should choose ¯ M = GN.

(7.190)

Because N¯ is a scalar, M is fixed to within a constant factor. Note that with this choice of M, we can write the controller equations as ¯ u = −Kxˆ + Nr, x˙ˆ = (F − LH)ˆx + Gu + Ly,

(7.191a) (7.191b)

which matches the configuration in Fig. 7.49(b). The net effect of this choice is that the control is computed from the feedback gain and the reference input before it is applied, and then the same control is input to both the plant and the estimator. In this form, if the plant control is subject to saturation (as shown by the inclusion of the saturation nonlinearity in Fig. 7.49(b), and discussed in Chapter 9), the same control limits can be applied in Eq. (7.191) to the control entering the equation for the estimate xˆ , and the nonlinearity cancels out of the x˜ equation. This behavior is essential for proper estimator performance. The block diagram corresponding to this technique is shown in Fig. 7.49(b). We will return to the selection of the gain factor on the reference ¯ in Section 7.9.2 after discussing the other two methods of selecting M. input, N, CASE 2. The second approach suggested earlier is to use the tracking error. This solution is sometimes forced on the control designer when the sensor measures only the output error. For example, in many thermostats the output is the difference between the temperature to be controlled and the setpoint temperature, and there is no absolute indication of the reference temperature available to the controller. Also, some radar tracking systems have a reading that is proportional to the pointing error, and this error signal alone must be used for feedback control. In these situations, we must select M and N¯ so that Eqs. (7.188) are driven by the error only. This requirement is satisfied if we select N¯ = 0 and M = −L. (7.192) Then the estimator equation is x˙ˆ = (F − GK − LH)ˆx + L(y − r).

(7.193)

The compensator in this case, for low-order designs, is a standard lead compensator in the forward path. As we have seen in earlier chapters, this design can have a considerable amount of overshoot because of the zero of the compensator. This design corresponds exactly to the compensators designed by the transform methods given in Chapters 5 and 6. CASE 3. The third method of selecting M and N¯ is to choose the values so as to assign the system’s zeros to arbitrary locations of the designer’s choice. This method provides the designer with the maximum flexibility in satisfying transient-response and steady-state gain constraints. The other two methods are special cases of this third method. All three methods depend on the zeros. As we saw in Section 7.5.2,

7.9 Introduction of the Reference Input with the Estimator

495

when there is no estimator and the reference input is added to the control, the closedloop system zeros remain fixed as the zeros of the open-loop plant. We now examine what happens to the zeros when an estimator is present. To do so, we reconsider the controller of Eqs. (7.188). If there is a zero of transmission from r to u, then there is necessarily a zero of transmission from r to y, unless there is a pole at the same location as the zero. It is therefore sufficient to treat the controller alone to determine what effect the choices of M and N¯ will have on the system zeros. The equations for a zero from r to u from Eqs. (7.188) are given by   sI − F + GK + LH −M det = 0. (7.194) −K N¯ (We let y = 0 because we care only about the effect of r.) If we divide the last column by the (nonzero) scalar N¯ and then add to the rest the product of K times the last column, we find that the feed-forward zeros are at the values of s such that   K −M sI − F + GK + LH − M N¯ N¯ = 0, det 0 1 or

  M det sI − F + GK + LH − K = γ (s) = 0. N¯

(7.195)

Now Eq. (7.195) is exactly in the form of Eq. (7.136) for selecting L to yield desired locations for the estimator poles. Here we have to select M/N¯ for a desired zero polynomial γ (s) in the transfer function from the reference input to the control. Thus the selection of M provides a substantial amount of freedom to influence the transient response. We can add an arbitrary nth-order polynomial to the transfer function from r to u and hence from r to y; that is, we can assign n zeros in addition to all the poles that we assigned previously. If the roots of γ (s) are not canceled by the poles of the system, then they will be included in zeros of transmission from r to y. ¯ Two considerations can guide us in the choice of M/N—that is, in the location of the zeros. The first is dynamic response. We have seen in Chapter 3 that the zeros influence the transient response significantly, and the heuristic guidelines given there may suggest useful locations for the available zeros. The second consideration, which will connect state-space design to another result from transform techniques, is steadystate error or velocity-constant control. In Chapter 4 we derived the relationship between the steady-state accuracy of a Type 1 system and the closed-loop poles and zeros. If the system is Type 1, then the steady-state error to a step input will be zero and to a unit-ramp input will be 1 e∞ = , (7.196) Kv

Truxal’s formula

where Kv is the velocity constant. Furthermore, it was shown that if the closed-loop poles are at {pi } and the closed-loop zeros are at {zi }, then (for a Type 1 system) Truxal’s formula gives 1  1 1 = − . (7.197) Kv zi pi

496 Chapter 7 State-Space Design Equation (7.197) forms the basis for a partial selection of γ (s), and hence of M and ¯ The choice is based on two observations: N. 1. If |zi − pi | 1, then the effect of this pole–zero pair on the dynamic response will be small, because the pole is almost canceled by the zero, and in any transient the residue of the pole at pi will be very small. 2. Even though zi − pi is small, it is possible for 1/zi − 1/pi to be substantial and thus to have a significant influence on Kv according to Eq. (7.197). Application of these two guidelines to the selection of γ (s), and hence of M and ¯ results in a lag-network design. We illustrate this with an example. N,

EXAMPLE 7.34

Servomechanism: Increasing the Velocity Constant through Zero Assignment

Lag compensation by a state-space method

Consider the second-order servomechanism system described by 1 G(s) = s(s + 1) and with state description x˙ 1 = x2 , x˙ 2 = −x2 + u. Design a controller using pole placement so that both poles are at s = −2 and the system has a velocity constant Kv = 10. Derive an equivalent discrete controller with a sampling period of Ts = 0.1 sec (20 × ωn = 20 × 0.05 = 0.1 sec), and compare the continuous and digital control outputs, as well as the control efforts. Solution. For this problem, the state feedback gain K=[ 8 3 ] results in the desired control poles. However, with this gain, Kv = 2, and we need Kv = 10. What effect will using estimators designed according to the three methods for M and N¯ selection have on our design? Using the first strategy (the autonomous estimator), we find that the value of Kv does not change. If we use the second method (error control), we introduce a zero at a location unknown beforehand, and the effect on Kv will not be under direct design control. However, if we use the third option (zero placement) along with Truxal’s formula [Eq. (7.197)], we can satisfy both the dynamic response and the steady-state requirements. First we must select the estimator pole p3 and the zero z3 to satisfy Eq. (7.197) for Kv = 10. We want to keep z3 −p3 small, so that there is little effect on the dynamic response, and yet have 1/z3 − 1/p3 be large enough to increase the value of Kv . To do this, we arbitrarily set p3 small compared with the control dynamics. For example, we let p3 = −0.1. Notice that this approach is opposite to the usual philosophy of estimation design, where fast response is the requirement. Now, using Eq. (7.197) to get 1 1 1 1 1 = − − − , Kv z3 p1 p2 p3 where p1 = −2+2j, p2 = −2−2j, and p3 = −0.1, we solve for z3 such that Kv = 10 we obtain

7.9 Introduction of the Reference Input with the Estimator

497

1 1 1 4 1 + , = + = Kv 8 0.1 z3 10

or

1 = −0.096. 10.4 We thus design a reduced-order estimator to have a pole at −0.1 and choose M/N¯ such that γ (s) has a zero at −0.096. A block diagram of the resulting system is shown in Fig. 7.50(a). You can readily verify that this system has the overall transfer function 8.32(s + 0.096) Y (s) = 2 , (7.198) R(s) (s + 4s + 8)(s + 0.1) for which Kv = 10, as specified. The compensation shown in Fig. 7.50(a) is nonclassical in the sense that it has two inputs (e and y) and one output. If we resolve the equations to provide pure error compensation by finding the transfer function from e and u, which would give Eq. (7.198), we obtain the system shown in Fig. 7.50(b). This can be seen as follows: The relevant controller equations are x˙ c = 0.8 e − 3.1 u, z3 = −

u = 8.32 e + 3.02 y + xc , where xc is the controller state. Taking the Laplace transform of these equations, eliminating Xc (s), and substituting for the output [Y (s) = G(s)U(s)], we find that the compensator is described by U(s) (s + 1)(8.32s + 0.8) = Dc (s) = . E(s) (s + 4.08)(s + 0.0196) This compensation is a classical lag–lead network. The root locus of the system in Fig. 7.50(b) is shown in Fig. 7.51. Note the pole–zero pattern near the origin that is Figure 7.50 Servomechanism with assigned zeros (a lag network): (a) the two-input compensator; (b) equivalent unity-feedback system

R






e

 0.8 




8.32

1 s

xc 






u



1 s(s  1)

Y

3.1 3.02

(a)

R

 




e

(s  1)(8.32s  0.8) (s  4.08)(s  0.0196)

(b)

u

1 s(s  1)

Y

498 Chapter 7 State-Space Design Figure 7.51 Root locus of lag–lead compensation

0.12 0.08

0.24

0.16

0.08

0.08 Im(s)

0.12 1 4

3

2

1

Re(s) 1

Figure 7.52 Frequency response of lag–lead compensation

1000

Magnitude

100 10 1 0.1 0.01 0.001 0.0001 0.01

0.1

0.4 1 2 4 6 10

40 100

v (rad/sec) (a)

Phase

90 120 150 180 0.01

0.1

0.4 1 2 4 6 10 v (rad/sec) (b)

40 100

7.9 Introduction of the Reference Input with the Estimator Figure 7.53 Step response of the system with lag compensation

499

y 1.2 1.0 0.8 0.6 0.4 0.2 0

MATLAB c2d

0

1

2 3 Time (sec)

4

5

characteristic of a lag network. The Bode plot in Fig. 7.52 shows the phase lag at low frequencies and the phase lead at high frequencies. The step response of the system is shown in Fig. 7.53 and shows the presence of a “tail” on the response due to the slow pole at −0.1. Of course, the system is Type 1 and the system will have zero tracking error eventually. The discrete equivalent for the controller is obtained from MATLAB by using the c2d command, as in the following code: nc=conv([1 1],[8.32 0.8]); % controller numerator dc=conv([1 4.08],[1 0.0196]); % controller denominator sysDc=tf(nc,dc); % form controller system description ts=0.1; % sampling time of 0.1 sec sysDd=c2d(sysDc,ts,’zoh’); % convert to discrete time controller The discrete controller has the discrete transfer function Dd (z) =

8.32(z − 0.9903)(z − 0.9191) 8.32z2 − 15.8855z + 7.5721 = . z2 − 1.6630z + 0.6637 (z − 0.998)(z − 0.6665)

The equation for the control law (with sample period suppressed for clarity) is u(k + 1) = 1.6630u(k) + 0.6637u(k − 1) + 8.32e(k + 1) − 15.8855e(k) + 7.5721e(k − 1). SIMULINK simulation

A SIMULINK diagram for simulating both the continuous and discrete systems is shown in Fig. 7.54. A comparison of the continuous and discrete step responses and control signals is shown in Fig. 7.55. Better agreement between the two responses can be achieved if the sampling period is reduced. ¯ this time to We now reconsider the first two methods for choosing M and N, examine their implications in terms of zeros. Under the first rule (for the autonomous ¯ Substituting this into Eq. (7.195) yields, for the controller estimator), we let M = GN. feed-forward zeros, det(sI − F + LH) = 0. (7.199) This is exactly the equation from which L was selected to make the characteristic polynomial of the estimator equation equal to αe (s). Thus we have created n zeros in

500 Chapter 7 State-Space Design   Step 1

Continuous controller

Plant

8.32s2  9.12s  0.8 s2  4.0996s  0.08

1 s2  s

Mux

 

Discrete controller

Plant1

8.32z2  15.8848z  7.5714 z2  1.6630z  0.6637

1 s2  s

Control

Mux1

Output

Figure 7.54 SIMULINK
block diagram to compare continuous and discrete controllers exactly the same locations as the n poles of the estimator. Because of this pole–zero cancellation (which causes “uncontrollability” of the estimator modes), the overall transfer function poles consist only of the state feedback controller poles. The second rule (for a tracking-error estimator) selects M = −L and N¯ = 0. If these are substituted into Eq. (7.194), then the feed-forward zeros are given by   sI − F + GK + LH L det = 0. (7.200) −K 0

Transfer function for the closed-loop system when reference input is included in controller

If we postmultiply the last column by H and subtract the result from the first n columns, and then premultiply the last row by G and add it to the first n rows, Eq. (7.200) then reduces to   sI − F L det = 0. (7.201) −K 0 If we compare Eq. (7.201) with the equations for the zeros of a system in a state description, Eq. (7.66), we see that the added zeros are those obtained by replacing the input matrix with L and the output with K. Thus, if we wish to use error control, we have to accept the presence of these compensator zeros that depend on the choice of K and L and over which we have no direct control. For low-order cases this results, as we said before, in a lead compensator as part of a unity feedback topology. Let us now summarize our findings on the effect of introducing the reference input. When the reference input signal is included in the controller, the overall transfer function of the closed-loop system is Ks γ (s)b(s) Y (s) = , (7.202) R(s) αe (s)αc (s) where Ks is the total system gain and γ (s) and b(s) are monic polynomials. The polynomial αc (s) results in a control gain K such that det[sI − F + GK] = αc (s). The polynomial αe (s) results in estimator gains L such that det[sI − F + LH] = αe (s). Because, as designers, we get to choose αc (s) and αe (s), we have complete freedom in assigning the poles of the closed-loop system. There are three ways to handle the T (s) =

7.9 Introduction of the Reference Input with the Estimator Figure 7.55 Comparison of step responses and control signals for continuous and discrete controllers: (a) step responses; (b) control signals

501

1.4 Digital controller 1.2 1.0 0.8 y

Continuous controller 0.6 0.4 0.2 0

0

1

2 3 Time (sec)

4

5

4

5

(a) 10 8 6 u

4 2 Continuous controller 0 Digital controller 2

0

1

2 3 Time (sec) (b)

polynomial γ (s): We can select it so that γ (s) = αe (s) by using the implementation of Fig. 7.49(b), in which case M/N¯ is given by Eq. (7.190); we may accept γ (s) as given by Eq. (7.201), so that error control is used; or we may give γ (s) arbitrary coefficients by selecting M/N¯ from Eq. (7.195). It is important to point out that the plant zeros represented by b(s) are not moved by this technique and remain as part of the closedloop transfer function unless αc or αe are selected to cancel some of these zeros.

7.9.2 Selecting the Gain We now turn to the process of determining the gain N¯ for the three methods of selecting M. If we choose method 1, the control is given by Eq. (7.191a) and xˆ ss = xss . Therefore, we can use either N¯ = Nu + KNx , as in Eq. (7.102), or u = Nu r − K(ˆx − Nx r). This is the most common choice. If we use the second method, the result is

502 Chapter 7 State-Space Design trivial; recall that N¯ = 0 for error control. If we use the third method, we pick N¯ such that the overall closed-loop DC gain is unity.11 The overall system equations then are        x˙ G F − GK GK x ¯ Nr, (7.203a) = + ¯ x˜ 0 F − LH G−M x˙˜   x y=[ H 0 ] , (7.203b) x˜ ¯ is the outcome of selecting zero locations with either Eq. (7.195) or where M Eq. (7.190). The closed-loop system has unity DC gain if   −1  G F − GK GK −[ H 0 ] N¯ = 1. (7.204) ¯ 0 F − LH G−M ¯ we get12 If we solve Eq. (7.204) for N, N¯ = −

1 H(F − GK)−1 G[1 − K(F − LH)−1 (G

¯ − M)]

.

(7.205)

The techniques in this section can be readily extended to reduced-order estimators.

7.10 Integral Control and Robust Tracking The choices of N¯ gain in Section 7.9 will result in zero steady-state error to a step command, but the result is not robust because any change in the plant parameters will cause the error to be nonzero. We need to use integral control to obtain robust tracking. In the state-space design methods discussed so far, no mention has been made of integral control, and no design examples have produced a compensation containing an integral term. In Section 7.10.1 we show how integral control can be introduced by a direct method of adding the integral of the system error to the equations of motion. Integral control is a special case of tracking a signal that does not go to zero in the steady-state. We introduce (in Section 7.10.2) a general method for robust tracking that will present the internal model principle, which solves an entire class of tracking problems and disturbance-rejection controls. Finally, in Section 7.10.3, we show that if the system has an estimator and also needs to reject a disturbance of known structure, we can include a model of the disturbance in the estimator equations and use the computer estimate of the disturbance to cancel the effects of the real plant disturbance on the output. 11A reasonable alternative is to select N ¯ such that, when r and y are both unchanging, the DC gain from r to u is the negative of the DC gain from y to u. The consequences of this choice are that our controller can be structured as a combination of error control and generalized derivative control, and if the system is capable of Type 1 behavior, that capability will be realized. 12 We have used the fact that



A 0

C B

−1

 =

A−1 0

−A−1 CB−1 B−1

 .

7.10 Integral Control and Robust Tracking

503

7.10.1 Integral Control We start with an ad hoc solution to integral control by augmenting the state vector with the desired dynamics. For the system x˙ = Fx + Gu + G1 w,

(7.206a)

y = Hx,

(7.206b)

e = y − r, as well as the state of the we can feed back the integral of the plant, x, by augmenting the plant state with the extra (integral) state xI , which obeys the differential equation x˙ I = Hx − r(= e). Thus

error,13

xI =

t

e dt.

Augmented state equations with integral control

The augmented state equations become            0 1 0 0 H xI x˙ I + u− r+ = w, x G 0 G1 x˙ 0 F

Feedback law with integral control

and the feedback law is

 u = −[ K1

or simply

K0 ]

xI x

(7.207)

 ,



 xI . x With this revised definition of the system, we can apply the design techniques from Section 7.5 in a similar fashion; they will result in the control structure shown in Fig. 7.56. u = −K

EXAMPLE 7.35

Integral Control of a Motor Speed System Consider the motor speed system described by 1 Y (s) = ; U(s) s+3 that is, F = −3, G = 1, and H = 1. Design the system to have integral control and two poles at s = −5. Design an estimator with pole at s = −10. The disturbance enters at the same place as the control. Evaluate the tracking and disturbance rejection responses.

Figure 7.56 Integral control structure

r






e

1 s

xI

K1



Plant




u



y x

K0

13 Watch out for the sign here; we are using the negative of the usual convention.

504 Chapter 7 State-Space Design Solution. The pole-placement requirement is equivalent to pc = [−5; −5]. The augmented system description, including the disturbance w, is          0 1 xI 0 1 x˙ I = + (u + w) − r. x˙ 0 −3 x 1 0 Therefore, we can find K from       0 1 0 det sI − + K = s2 + 10s + 25, 0 −3 1 or s2 + (3 + K0 )s + K1 = s2 + 10s + 25. Consequently, K = [ K1

K0 ] = [ 25

7 ].

We may verify this result by using acker. The system is shown with feedbacks in Fig. 7.57, along with a disturbance input w. The estimator gain L = 7 is obtained from αe (s) = s + 10 = s + 3 + L. The estimator equation is of the form x˙ˆ = (F − LH)ˆx + Gu + Ly = −10ˆx + u + 7y, and u = −K0 xˆ = −7ˆx . The step response y1 due to a step reference input r and the output disturbance response y2 due to a step disturbance input w are shown in Fig. 7.58(a) and the associated control efforts (u1 and u2 ) are shown in Fig. 7.58(b). As expected, the system is Type 1 and tracks the step reference input and rejects the step disturbance asymptotically.

Figure 7.57 Integral control example

Plant w r






e

1 s

xI

25















1 s3

7 Estimator xˆ

y

7.10 Integral Control and Robust Tracking Figure 7.58 Transient response for motor speed system: (a) step responses; (b) control efforts

505

1 0.9

y1

0.8 0.7 0.6 y 0.5 0.4 0.3 0.2 0.1 0

y2 0

0.5

1

1.5

2 2.5 3 Time (sec)

3.5

4

4.5

5

3.5

4

4.5

5

(a) 3.5 u1 3 2.5 2 u(t) 1.5 1 0.5

u2

0 0.5

0

0.5

1

1.5

2 2.5 3 Time (sec) (b)


7.10.2 Robust Tracking Control: The Error-Space Approach In Section 7.10.1 we introduced integral control in a direct way and selected the structure of the implementation so as to achieve integral action with respect to reference and disturbance inputs. We now present a more analytical approach to giving a control system the ability to track (with zero steady-state error) a nondecaying input and to reject (with zero steady-state error) a nondecaying disturbance such as a step, ramp, or sinusoidal input. The method is based on including the equations satisfied by these external signals as part of the problem formulation and solving the problem of control in an error space, so we are assured that the error approaches zero even if the output is following a nondecaying, or even a growing, command (such as a ramp signal) and even if some parameters change (the robustness property). The method is illustrated in detail for signals that satisfy differential equations of order 2, but the extension to more complex signals is not difficult.

506 Chapter 7 State-Space Design Suppose we have the system state equations x˙ = Fx + Gu + G1 w,

(7.208a)

y = Hx,

(7.208b)

and a reference signal that is known to satisfy a specific differential equation. The initial conditions on the equation generating the input are unknown. For example, the input could be a ramp whose slope and initial value are unknown. Plant disturbances of the same class may also be present. We wish to design a controller for this system so that the closed-loop system will have specified poles, and can also track input command signals, and reject disturbances of the type described without steady-state error. We will develop the results only for second-order differential equations. We define the reference input to satisfy the relation r¨ + α1 r˙ + α2 r = 0

(7.209)

and the disturbance to satisfy exactly the same equation: w ¨ + α1 w˙ + α2 w = 0.

(7.210)

The (tracking) error is defined as e = y − r. The meaning of robust control

(7.211)

The problem of tracking r and rejecting w can be seen as an exercise in designing a control law to provide regulation of the error, which is to say that the error e tends to zero as time gets large. The control must also be structurally stable or robust, in the sense that regulation of e to zero in the steady-state occurs even in the presence of “small” perturbations of the original system parameters. Note that, in practice, we never have a perfect model of the plant, and the values of parameters are virtually always subject to some change, so robustness is always very important. We know that the command input satisfies Eq. (7.209), and we would like to eliminate the reference from the equations in favor of the error. We begin by replacing r in Eq. (7.209) with the error of Eq. (7.211). When we do this, the reference cancels because of Eq. (7.209), and we have the formula for the error in terms of the state: e¨ + α1 e˙ + α2 e = y¨ + α1 y˙ + α2 y = H¨x + α1 Hx˙ + α2 Hx.

(7.212a) (7.212b)

We now replace the plant state vector with the error-space state, defined by ξ  x¨ + α1 x˙ + α2 x.

(7.213)

Similarly, we replace the control with the control in error space, defined as µ  u¨ + α1 u˙ + α2 u.

(7.214)

With these definitions, we can replace Eq. (7.212b) with e¨ + α1 e˙ + α2 e = Hξ . Robust control equations in the error space

The state equation for ξ is given ... ξ˙ = x + α1 x¨ + α2 x˙ = Fξ + Gµ.

(7.215)

by14

(7.216)

14 Notice that this concept can be extended to more complex equations in r and to multivariable systems.

7.10 Integral Control and Robust Tracking

507

Notice that the disturbance, as well as the reference, cancels from Eq. (7.216). Equations (7.215) and (7.216) now describe the overall system in an error space. In standard state-variable form, the equations are z˙ = Az + Bµ, where z = [ e e˙

ξ T ]T and ⎡ 0 A = ⎣ −α2 0

1 −α1 0

⎤ 0 H ⎦, F

(7.217) ⎡

⎤ 0 B = ⎣ 0 ⎦. G

(7.218)

The error system (A, B) can be given arbitrary dynamics by state feedback if it is controllable. If the plant (F, G) is controllable and does not have a zero at any of the roots of the reference-signal characteristic equation αr (s) = s2 + α1 s + α2 , then the error system (A, B) is controllable.15 We assume these conditions hold; therefore, there exists a control law of the form ⎡ ⎤ e (7.219) µ = −[ K2 K1 K0 ] ⎣ e˙ ⎦ = −Kz, ξ such that the error system has arbitrary dynamics by pole placement. We now need to express this control law in terms of the actual process state x and the actual control. We combine Eqs. (7.219), (7.213), and (7.214) to get the control law in terms of u 2 and x (we write u(2) to mean ddt 2u ): (u + K0 x)(2) +

2 

αi (u + K0 x)(2−i) = −

i=1

2 

Ki e(2−i) .

(7.220)

i=1

The structure for implementing Eq. (7.220) is very simple for tracking constant inputs. In that case the equation for the reference input is r˙ = 0. In terms of u and x, the control law [Eq. (7.220)] reduces to u˙ + K0 x˙ = −K1 e.

(7.221)

Here we need only to integrate to reveal the control law and the action of integral control:

t u = −K1 e dτ − K0 x. (7.222) A block diagram of the system, shown in Fig. 7.59, clearly shows the presence of a pure integrator in the controller. In this case the only difference between the internal model method of Fig. 7.59 and the ad hoc method of Fig. 7.57 is the relative location of the integrator and the gain. A more complex problem that clearly shows the power of the error-space approach to robust tracking is posed by requiring that a sinusoid be tracked with zero steadystate error. The problem arises, for instance, in the control of a mass-storage disk-head assembly. 15 For example, it is not possible to add integral control to a plant that has a zero at the origin.

508 Chapter 7 State-Space Design Figure 7.59 Integral control using the internal model approach

r






e

K1

1 s



Plant




u



y x

K0

EXAMPLE 7.36

Disk-Drive Servomechanism: Robust Control to Follow a Sinusoid A simple normalized model of a computer disk-drive servomechanism is given by the equations     0 1 0 F= , G= , 0 −1 1    0 G1 = , H = 1 0 , J = 0. 1 Because the data on the disk are not exactly on a centered circle, the servo must follow a sinusoid of radian frequency ω0 determined by the spindle speed. (a) Give the structure of a controller for this system that will follow the given reference input with zero steady-state error. √ (b) Assume ω0 = 1 and that the desired closed-loop poles are at −1 ± j 3 and √ − 3 ± j1. (c) Demonstrate the tracking and disturbance rejection properties of the system using MATLAB or SIMULINK. Solution (a) The reference input satisfies the differential equation r¨ = −ω02 r so that α1 = 0 and α2 = ω02 . With these values, the error-state matrices, according to Eq. (7.218), are ⎤ ⎤ ⎡ ⎡ 0 1 0 0 0 ⎥ ⎢ ⎢ −ω2 0 1 0 ⎥ 0 ⎥, B = ⎢ 0 ⎥. A=⎢ ⎦ ⎦ ⎣ ⎣ 0 0 0 0 1 1 0 0 0 −1 The characteristic equation of A − BK is s4 + (1 + K02 )s3 + (ω02 + K01 )s2 + [K1 + ω02 (1 + K02 )]s + K01 ω02 K2 = 0,

Internal model principle

from which the gain may be selected by pole assignment. The compensator implementation from Eq. (7.220) has the structure shown in Fig. 7.60, which clearly shows the presence of the oscillator with frequency ω0 (known as the internal model of the input generator) in the controller.16 16 This is a particular case of the internal model principle, which requires that a model of the external or

exogenous signal be in the controller for robust tracking and disturbance rejection.

7.10 Integral Control and Robust Tracking Figure 7.60 Structure of the compensator for the servomechanism to track exactly the sinusoid of frequency ω0

r



509

Compensator




e

 K2

K1





1 s




Plant





1 s




1 s(s  1)

u




v02






K01



Internal model

K02

y

x1

x2

(b) Now assume that ω0 = 1 rad/sec and the desired closed-loop poles are as given: √ √ √ √ pc = [−1 + j ∗ 3; −1 − j ∗ 3; − 3 + j; − 3 − j]. Then the feedback gain is K = [K2 K1 : K0 ] = [2.0718 16.3923 : 13.9282 4.4641], which results in the controller x˙ c = Ac xc + Bc e, u = Cc xc , with

 Ac = Cc =



0 −1 1

0

1 0



 ,

Bc =

−16.3923 −2.0718

.

The relevant MATLAB statements are % plant matrices F=[0 1; 0 −1]; G=[0;1]; H=[1 0]; J=[0]; % form error space matrices omega=1; A=[0 1 0 0;−omega*omega 0 1 0;0 0 0 1;0 0 0 −1]; B=[0;0;G];

 ,

510 Chapter 7 State-Space Design % desired closed-loop poles pc=[−1+sqrt(3)*j ;−1−sqrt(3)*j;−sqrt(3)+j;−sqrt(3)−j]; K=place(A,B,pc); % form controller matrices K1=K(:,1:2); Ko=K(:,3:4); Ac=[0 1;−omega*omega 0]; Bc=−[K(2);K(1)];; Cc=[1 0]; Dc=[0]; The controller frequency response is shown in Fig. 7.61 and shows a gain of infinity at the rotation frequency of ω0 = 1 rad/sec. The frequency response from r to e[i.e., the sensitivity function S(s)], is shown in Fig. 7.62 and reveals a sharp notch at the rotation frequency ω0 = 1 rad/sec. The same notch is also present in the frequency response of the transfer function from w to y. (c) Figure 7.63 shows the SIMULINK simulation diagram for the system. Although the simulations can also be done in MATLAB, it is more instructive to use the interactive graphical environment of SIMULINK. SIMULINK also provides the capability to add nonlinearities (see Chapter 9) and carry out robustness studies efficiently.17 The tracking properties of the system are shown in Fig. 7.64(a), showing the asymptotic tracking property of the system. The associated control effort and the tracking error signal are shown in Fig. 7.64(b) and (c) respectively. The disturbance rejection properties of the system are illustrated in Fig. 7.65(a), displaying asymptotic disturbance rejection of sinusoidal disturbance input. The associated control effort is shown in Fig. 7.65(b). The closed-loop frequency

Magnitude (db)

250 200 150 100 50 0 102

101

100

101

100

101

v (rad/sec)

Phase (deg)

Figure 7.61 Controller frequency response

100 50 0 50 100 150 200 102

101 v (rad/sec)

17 In general, the design can be done in MATLAB and (nonlinear) simulations can be carried out in

SIMULINK.

7.10 Integral Control and Robust Tracking Figure 7.62 Sensitivity function frequency response

511

Magnitude (db)

50 0 50 100 150 200 101

100

101

102

101

102

v (rad/sec)

Phase (deg)

300 250 200 150 100 101

100 v (rad/sec)

To workspace

To workspace2

To workspace1

r

e

y

Scope2

Scope1

1 Integrator1

r

 

e

16.392 Integrator Gain

2.0718 Gain1

 

1 s

Integrator2  

1 s

u

1 s

 

1 Gain5

  1 Gain4

 

4.4641 Gain2 13.928 Gain3

Figure 7.63 SIMULINK
block diagram for robust servomechanism

Integrator3 x2 1 x1 s

Out 1

Scope

512 Chapter 7 State-Space Design 1.5 1 Reference, output

Figure 7.64 (a) Tracking properties for robust servomechanism; (b) control effort; (c) tracking error signal

r

0.5 y 0 0.5 1 1.5

0

5

10 15 (a) Time (sec)

20

25

0

5

10 15 (b) Time (sec)

20

25

0

5

10 15 (c) Time (sec)

20

25

1.5 1

Control, u

0.5 0 −0.5 −1 −1.5 −2

0.7 0.6

Error signal, e

0.5 0.4 0.3 0.2 0.1 0 −0.1

SIMULINK simulation

response [i.e., the complementary transfer function T (s)] for the robust servomechanism is shown in Fig. 7.66. As seen from the figure, the frequency response from r to y is unity at ω0 = 1 rad/sec as expected. The zeros of the system from r to e are located at ±j, −2.7321 ± j2.5425. The robust tracking properties are due to the presence of the blocking zeros at

7.10 Integral Control and Robust Tracking 1

w

0.8 0.6 Disturbance, output

Figure 7.65 (a) Disturbance rejection properties for robust servomechanism; (b) control effort

0.4 0.2

y

0 ⫺0.1 ⫺0.4 ⫺0.6 ⫺0.8 ⫺1

0

5

10 15 (a) Time (sec)

20

25

0

5

10 15 (b) Time (sec)

20

25

1

Control, u

0.5 0 −0.5 −1 −1.5

Magnitude (db)

10 0 10 20 30 40 102

101

100

101

100

101

v (rad/sec)

Phase (deg)

Figure 7.66 Closed-loop frequency response for robust servomechanism

90 45 0 45 90 135 180 225 270 102

101 v (rad/sec)

513

514 Chapter 7 State-Space Design Blocking zeros

±j. The zeros from w to y, both blocking zeros, are located at ±j. The robust disturbance rejection properties are due to the presence of these blocking zeros. From the nature of the pole-placement problem, the state z in Eq. (7.217) will tend toward zero for all perturbations in the system parameters as long as A − BK remains stable. Notice that the signals that are rejected are those that satisfy the equations with the values of αi actually implemented in the model of the external signals. The method assumes that these are known and implemented exactly. If the implemented values are in error, then a steady-state error will result. Now let us repeat the example of Section 7.10.1 for integral control.

EXAMPLE 7.37

Integral Control Using the Error-Space Design For the system H(s) =

1 s+3

with the state-variable description F = −3,

G = 1,

H = 1,

construct a controller with poles at s = −5 to track an input that satisfies r˙ = 0. Solution. The error system is    e˙ 0 = z˙ 0

1 −3



e z



 +

0 1

 µ.

If we take the desired characteristic equation to be αc (s) = s2 + 10s + 25, then the pole-placement equation for K is det[sI − A + BK] = αc (s).

(7.223)

In detail, Eq. (7.223) is s2 + (3 + K0 )s + K1 = s2 + 10s + 25, which gives K = [ 25

7 ] = [ K1

K0 ],

and the system is implemented as shown in Fig. 7.67. The transfer function from r to e for this system, the sensitivity function s(s + 10) E(s) = S(s) = − 2 , R(s) s + 10s + 25 shows a blocking zero at s = 0, which prevents the constant input from affecting the error. The closed-loop transfer function—that is, the complementary sensitivity function—is

7.10 Integral Control and Robust Tracking Figure 7.67 Example of internal model with feed-forward

N

R






25



1 s










1 s3






Y

7

N

Figure 7.68 Internal model as integral control with feed-forward

515

N

R








e

K1

1 s















Plant




u



N

Y x

K0

25 Y (s) = T (s) = 2 . R(s) s + 10s + 25 The structure of Fig. 7.68 permits us to add a feed-forward of the reference input, which provides one extra degree of freedom in zero assignment. If we add a term proportional to r in Eq. (7.222), then

t u = −K1 e(τ ) dτ − K0 x + Nr. (7.224) This relationship has the effect of creating a zero at −K1 /N. The location of this zero can be chosen to improve the transient response of the system. For actual implementation, we can rewrite Eq. (7.224) in terms of e to get

t e(τ ) dτ − K0 x + N(y − e). (7.225) u = −K1 The block diagram for the system is shown in Fig. 7.68. For our example, the overall transfer function now becomes Y (s) Ns + 25 = 2 . R(s) s + 10s + 25 Notice that the DC gain is unity for any value of N and that, through our choice of N, we can place the zero at any real value to improve the dynamic response. A natural strategy for locating the zero is to have it cancel one of the system poles, in this case at s = −5. The step response of the system is shown in Fig. 7.69 for N = 5, as well as for N = 0 and 8. With the understanding that one pole can be cancelled in integral control designs, we make sure to choose one of the desired control poles such that it is both real and able to be cancelled through the proper choice of N.

516 Chapter 7 State-Space Design Figure 7.69 Step responses with integral control and feed-forward

1.2 N8 1.0 5 0

Amplitude

0.8 0.6 0.4 0.2 0

0

0.5

1.0

1.5 Time (sec)

2.0

2.5

3.0


7.10.3 The Extended Estimator Our discussion of robust control so far has used a control based on full-state feedback. If the state is not available, then as in the regular case, the full-state feedback, Kx, can be replaced by the estimates, Kxˆ , where the estimator is built as before. As a final look at ways to design control with external inputs, in this section we develop a method for tracking a reference input and rejecting disturbances. The method is based on augmenting the estimator to include estimates from external signals in a way that permits us to cancel out their effects on the system error. Suppose the plant is described by the equations x˙ = Fx + Gu + Gw,

(7.226a)

y = Hx,

(7.226b)

e = Hx − r.

(7.226c)

Furthermore, assume that both the reference r and the disturbance w are known to satisfy the equations18 αw (s)w = αρ (s)w = 0,

(7.227)

αr (s)r = αρ (s)r = 0,

(7.228)

where αρ (s) = s2 + α1 s + α2 , corresponding to polynomials αw (s) and αr (s) in Fig. 7.70(a). In general, we would select the equivalent disturbance polynomial αρ (s) in Fig. 7.70(b) to be the least common multiple of αw (s) and αr (s). The first step is to recognize that, as far as the steady-state response of the output is concerned, there is an input-equivalent signal ρ that satisfies the same equation as r and w and enters the system at the same place as the control signal, as shown in Fig. 7.70(b). As before, we must assume that the 18Again we develop the results for a second-order equation in the external signals; the discussion can be extended to higher-order equations.

7.10 Integral Control and Robust Tracking W

r

1 aw(s)

1 ar(s)



u 




 G(s)






1 ar(s) r 

u  

e




K

G(s)

e





xˆ

(a)

rˆ




Extended estimator

K (b)

r 

u



rˆ






W

517




 G(s)






e

Extended estimator xˆ

(c)

Figure 7.70 Block diagram of a system for tracking and disturbance rejection with extended estimator: (a) equivalent disturbance; (b) block diagram for design; (c) block diagram for implementation plant does not have a zero at any of the roots of Eq. (7.227). For our purposes here, we can replace Eqs. (7.226) with x˙ = Fx + G(u + ρ),

(7.229a)

e = Hx.

(7.229b)

If we can estimate this equivalent input, we can add to the control a term −ρˆ that will cancel out the effects of the real disturbance and reference and cause the output to track r in the steady-state. To do this, we combine Eqs. (7.226) and (7.227) into a state description to get

where z = [ρ

ρ˙

z˙ = Az + Bu,

(7.230a)

e = Cz,

(7.230b)

xT ]T . The matrices are ⎡ ⎤ 0 1 0 A = ⎣−α2 −α1 0 ⎦ , G 0 F C = [0

0

H].

⎡ ⎤ 0 B = ⎣0⎦, G

(7.231a) (7.231b)

The system given by Eqs. (7.231) is not controllable since we cannot influence ρ from u. However, if F and H are observable and if the system (F, G, H) does not have a zero that is also a root of Eq. (7.227), then the system of Eq. (7.231) will be observable, and we can construct an observer that will compute estimates of both the state of the plant and of ρ. The estimator equations are standard, but the control is not: z˙ˆ = Aˆz + Bu + L(e − Cˆz), (7.232a) u = − Kxˆ − ρ. ˆ

(7.232b)

In terms of the original variables, the estimator equations are ⎤ ⎡ ⎤ ⎡ ⎡ ˙ ⎤ ⎡ ⎤ ⎤⎡ ρˆ ρˆ 0 1 0 0 l1 ⎣ ρ¨ˆ ⎦ = ⎣ −α2 −α1 0 ⎦ ⎣ ρ˙ˆ ⎦ + ⎣ 0 ⎦ u + ⎣ l2 ⎦ [e − Hxˆ ]. (7.233) G L3 G 0 F xˆ x˙ˆ

518 Chapter 7 State-Space Design The overall block diagram of the system for design is shown in Fig. 7.70(b). If we write out the last equation for xˆ in Eq. (7.233) and substitute Eq. (7.232b), a simplification of sorts results because a term in ρˆ cancels out: x˙ˆ = Gρˆ + Fˆx + G(−Kxˆ − ρ) ˆ + L3 (e − Hxˆ ) = Fˆx + G(−Kxˆ ) + L3 (e − Hxˆ ) = Fˆx + G¯u + L3 (e − Hxˆ ). With the estimator of Eq. (7.233) and the control of Eq. (7.232b), the state equation is x˙ = Fx + G(−Kxˆ − ρ) ˆ + Gρ.

(7.234)

In terms of the estimate errors, Eq. (7.234) can be rewritten as x˙ = (F − GK)x + GKx˜ + Gρ. ˜

(7.235)

Because we designed the estimator to be stable, the values of ρ˜ and x˜ go to zero in the steady state, and the final value of the state is not affected by the external input. The block diagram of the system for implementation is drawn in Fig. 7.70(c). A very simple example will illustrate the steps in this process.

EXAMPLE 7.38

Steady-State Tracking and Disturbance Rejection of Motor Speed by Extended Estimator Construct an estimator to control the state and cancel a constant bias at the output and track a constant reference in the motor speed system described by x˙ = −3x + u,

(7.236a)

y = x + w,

(7.236b)

w˙ = 0,

(7.236c)

r˙ = 0.

(7.236d)

Place the control pole at s = −5 and the two extended estimator poles at s = −15. Solution. To begin, we design the control law by ignoring the equivalent disturbance. Rather, we notice by inspection that a gain of −2 will move the single pole from −3 to the desired −5, Therefore, K = 2. The system augmented with equivalent external input ρ, which replaces the actual disturbance w and the reference r, is given by ρ˙ = 0, x˙ = −3x + u + ρ, e = x. The extended estimator equations are ρˆ˙ = l1 (e − xˆ ), x˙ˆ = −3ˆx + u + ρˆ + l2 (e − xˆ ). The estimator error gain is found to be L = [ 225 27 ]T from the characteristic equation   s l1 det = s2 + 30s + 225. 1 s + 3 + l2

7.11 Loop Transfer Recovery (LTR) 2 1 W

r

0

u








 2

rˆ

 1 s3 







Extended estimator

e

y, x, ˆ rˆ

1 2 3

(a)

y xˆ rˆ

4 5 6 7

xˆ

519

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2 Time (sec) (b)

Figure 7.71 Motor speed system with extended estimator: (a) block diagram; (b) command step response and disturbance step response A block diagram of the system is given in Fig. 7.71(a), and the step responses to input at the command r (applied at t = 0 sec) and at the disturbance w (applied at t = 0.5 sec) are shown in Fig. 7.71(b).


7.11 Loop Transfer Recovery (LTR)

LTR

The introduction of an estimator in a state feedback controller loop may adversely affect the stability robustness properties of the system [i.e., the phase margin (PM) and gain margin (GM) properties may become arbitrarily poor, as shown by Doyle’s famous example (Doyle, 1978)]. However, it is possible to modify the estimator design so as to try to “recover” the LQR stability robustness properties to some extent. This process, called loop transfer recovery (LTR), is especially effective for minimumphase systems. To achieve the recovery, some of the estimator poles are placed at (or near) the zeros of the plant and the remaining poles are moved (sufficiently far) into the LHP. The idea behind LTR is to redesign the estimator in such a way as to shape the loop gain properties to approximate those of LQR. The use of LTR means that feedback controllers can be designed to achieve desired sensitivity [S(s)] and complementary sensitivity functions [T (s)] at critical (loop-breaking) points in the feedback system (e.g., at either the input or output of the plant). Of course, there is a price to be paid for this improvement in stability robustness! The newly designed control system may have worse sensor noise sensitivity properties. Intuitively, one can think of making (some of) the estimator poles arbitrarily fast so that the loop gain is approximately that of LQR. Alternatively, one can think of essentially “inverting” the plant transfer function so that all the LHP poles of the plant are cancelled by the dynamic compensator to achieve the desired loop shape. There are obvious trade-offs, and the designer needs to be careful to make the correct choice for the given problem, depending on the control system specifications.

520 Chapter 7 State-Space Design LTR is a well-known technique now, and specific practical design procedures have been identified (Athans, 1986; Stein and Athans, 1987; Saberi et al., 1993). The same procedures may also be applied to nonminimum phase systems, but there is no guarantee on the extent of possible recovery. The LTR technique may be viewed as a systematic procedure to study design trade-offs for linear quadraticbased compensator design (Doyle and Stein, 1981). We will now formulate the LTR problem. Consider the linear system x˙ = Fx + Gu + w,

(7.237a)

y = Hx + v,

(7.237b)

where w and v are uncorrelated zero-mean white Gaussian process and sensor noise with covariance matrices Rw ≥ 0 and Rv ≥ 0. The estimator design yields x˙ˆ = Fˆx + Gu + L(y − yˆ ),

(7.238a)

yˆ = Hxˆ ,

(7.238b)

resulting in the usual dynamic compensator Dc (s) = −K(sI − F + GK + LH)−1 L.

(7.239)

We will now treat the noise parameters, Rw and Rv , as design “knobs” in the dynamic compensator design. Without loss of generality, let us choose Rw =

T and Rv = 1. For LTR, assume that  = qG, where q is a scalar design parameter. The estimator design is then based on the specific design parameters Rw and Rv . It can be shown that, for a minimum-phase system, as q becomes large (Doyle and Stein, 1979), lim Dc (s)G(s) = K(sI − F)−1 G,

q→∞

Plant inversion

(7.240)

the convergence is pointwise in s and the degree of recovery can be arbitrarily good. This design procedure in effect “inverts” the plant transfer function in the limit as q → ∞: (7.241) lim Dc (s) = K(sI − F)−1 GG−1 (s). q→∞

This is precisely the reason that full-loop transfer recovery is not possible for a nonminimum-phase system. This limiting behavior may be explained using the symmetric root loci. As q → ∞, some of the estimator poles approach the zeros of Ge (s) = H(sI − F)−1
, LTR for nonminimum-phase systems

(7.242)

and the rest tend to infinity19 [see Eqs. (7.166) and (7.167)]. In practice, the LTR design procedure can still be applied to a nonminimum-phase plant. The degree of recovery will depend on the specific locations of the nonminimum-phase zeros. Sufficient recovery should be possible at many frequencies if the RHP zeros are located outside the specified closed-loop bandwidth. Limits on achievable performance of feedback systems due to RHP zeros are discussed in Freudenberg and Looze (1985). We will next illustrate the LTR procedure by a simple example. 19 In a Butterworth configuration.

7.11 Loop Transfer Recovery (LTR)

EXAMPLE 7.39

521

LTR Design for Satellite Attitude Control Consider the satellite system with state-space description     0 1 0 F= , G= , 0 0 1  H= 1 0 , J = 0. (a) Design an LQR controller with Q = ρHT H and R = 1, ρ = 1, and determine the loop gain. (b) Then design a compensator that recovers the LQR loop gain of part (a) using the LTR technique for q = 1, 10, 100. (c) Compare the different candidate designs in part (b) with respect to the actuator activity due to additive white Gaussian sensor noise. Solution. Using lqr, the selected LQR weights result in the feedback gain K = [1 1.414]. The loop transfer function is 1.414(s + 0.707) . s2 A magnitude frequency response plot of this LQR loop gain is shown in Fig. 7.72. For the estimator design using lqe, let  = qG, Rw =
T
, R v = 1, and choose q = 10, resulting in the estimator gain   14.142 L= . 100 K(sI − F)−1 G =

The compensator transfer function is Dc (s) = K(sI − F + GK + LH)−1 L 155.56(s + 0.6428) 155.56(s + 0.6428) = , = 2 (s + 15.556s + 121) (s + 7.77 + j7.77)(s + 7.77 − j7.77) and the loop transfer function is Dc (s)G(s) =

155.56(s + 0.6428) . s2 (s + 7.77 + j7.77)(s + 7.77 − j7.77)

Figure 7.72 shows the frequency response of the loop transfer function for several values of q (q = 1, 10,100), along with the ideal LQR loop transfer function frequency response. As seen from this figure, the loop gain tends to approach that of LQR as the value of q increases. As seen in Fig. 7.72, for q = 10, the “recovered” gain margin is GM = 11.1 = 20.9 db and the PM = 55.06◦ . Sample MATLAB statements to carry out the preceding LTR design procedure are as follows: F=[0 1; 0 0]; G=[0;1]; H=[1 0]; J=[0]; sys0=ss(F,G,H,J); H1=[1 0]; sys=ss(F,G,H1,J);

522 Chapter 7 State-Space Design GM Magnitude

Figure 7.72 Frequency response plots for LTR design

100

GM2 LQR

q  100 q  10 q1

GM1

105 101

100

101 v (rad/sec)

102

103

Phase (deg)

LQR 90 120 150 180 210 240 270 101

MATLAB lqr

MATLAB lqe

MATLAB bode MATLAB margin

PM

q  100

q  10

PM1 q1 100

101 v (rad/sec)

102

103

w=logspace(−1,3,1000); rho=1.0; Q=rho*H1’*H1; r=1; [K]=lqr(F,G,Q,r) sys1=ss(F,G,K,0); [maggk1,phasgk1,w]=bode(sys1,w); q=10; gam=q*G; Q1=gam
*gam; rv=1; [L]=lqe(F,gam,H,Q1,rv) aa=F−G*K−L*H; bb=L; cc=K; dd=0; sysk=ss(aa,bb,cc,dd); sysgk=series(sys0,sysk); [maggk,phsgk,w]=bode(sysgk,w); [gm,phm,wcg,wcp]=margin(maggk,phsgk,w) loglog(w,[maggk1(:) maggk(:)]); semilogx(w,[phasgk1(:) phsgk(:)]); To determine the effect of sensor noise, ν, on the actuator activity, we determine the transfer function from ν to u as shown in Fig. 7.73. For the selected value of LTR

7.11 Loop Transfer Recovery (LTR) Figure 7.73 Closed-loop system for LTR



R




Dc(s)

G(s)

523

Y








 V

Figure 7.74 SIMULINK block diagram for LTR  

State-space controller x  Ax  Bu y  Cx  Du

To workspace4 u

Scope2

Integrator1 1 s

Integrator 1 s  

Scope1 Band-limited white noise To workspace3 y

design parameter, q = 10, we have −Dc (s) −155.56s2 (s + 0.6428) U(s) = H(s) = = 4 . V (s) 1 + Dc (s)G(s) s + 15.556s3 + 121s2 + 155.56s + 99.994 RMS value

SIMULINK simulation

One reasonable measure of the effect of the sensor noise on the actuator activity is the root-mean-square (RMS) value of the control, u, due to the additive noise, ν. The RMS value of the control may be computed as  T0 1/2 1 urms = u(t)2 dt (7.243) T0 0 where T0 is the signal duration. Assuming white Gaussian noise ν, the RMS value of the control can also be determined analytically (Boyd and Barratt, 1991). The closed-loop SIMULINK diagram with band-limited white sensor noise excitation is shown in Fig. 7.74. The values of the RMS control were computed for different values of the LTR design parameter q, using the SIMULINK simulations, and are tabulated in Table 7.2. The results suggest increased vulnerability due to actuator wear as q is increased. Refer to MATLAB commands ltry and ltru for the LTR computations.

524 Chapter 7 State-Space Design

TABLE 7.2

Computed RMS Control for Various Values of LTR Tuning Parameter q. q 1 10 100

 u rms 0.1454 2.8054 70.5216


7.12 Direct Design with Rational Transfer Functions

General controller in polynomial form

An alternative to the state-space methods discussed so far is to postulate a generalstructure dynamic controller with two inputs (r and y) and one output (u) and to solve for the transfer function of the controller to give a specified overall r-to-y transfer function. A block diagram of the situation is shown in Fig. 7.75. We model the plant as the transfer function Y (s) b(s) = , (7.244) U(s) a(s) rather than by state equations. The controller is also modeled by its transfer function, in this case a transfer function with two inputs and one output: U(s) = −

cy (s) cr (s) Y (s) + R(s). d(s) d(s)

(7.245)

Here d(s), cy (s), and cr (s) are polynomials. In order for the controller of Fig. 7.75 and Eq. (7.245) to be implemented, the orders of the numerator polynomials cy (s) and cr (s) must not be higher than the order of the denominator polynomial d(s). To carry out the design, we require that the closed-loop transfer function defined by Eqs. (7.244) and (7.245) be matched to the desired transfer function Y (s) cr (s)b(s) = . R(s) αc (s)αe (s)

(7.246)

Equation (7.246) tells us that the zeros of the plant must be zeros of the overall system. The only way to change this is to have factors of b(s) appear in either αc or αe . We combine Eqs. (7.244) and (7.245) to get   cy (s) cr (s) a(s)Y (s) = b(s) − Y (s) + R(s) , (7.247) d(s) d(s) which can be rewritten as [a(s)d(s) + b(s)cy (s)]Y (s) = b(s)cr (s)R(s). Diophantine equation

(7.248)

Comparing Eq. (7.246) with Eq. (7.247) we immediately see that the design can be accomplished if we can solve the Diophantine equation a(s)d(s) + b(s)cy (s) = αc (s)αe (s)

(7.249)

for given arbitrary a, b, αc , and αe . Because each transfer function is a ratio of polynomials, we can assume that a(s) and d(s) are monic polynomials; that is, the coefficient of the highest power of s in each polynomial is unity. The question is, How

7.12 Direct Design with Rational Transfer Functions Figure 7.75 Direct transfer-function formulation

R(s)

cr(s)








1 d(s)

b(s) a(s)

525

Y(s)

cy(s) Dynamic controller

Dimension of the controller

many equations and how many unknowns are there if we match coefficients of equal powers of s in Eq. (7.249)? If a(s) is of degree n (given) and d(s) is of degree m (to be selected), then a direct count yields 2m + 1 unknowns in d(s) and cy (s) and n + m equations from the coefficients of powers of s. Thus the requirement is that 2m + 1 ≥ n + m, or m ≥ n − 1. One possibility for a solution is to choose d(s) of degree n and cy (s) of degree n−1. In that case, which corresponds to the state-space design for a full-order estimator, there are 2n equations and 2n unknowns with αc αe of degree 2n. The resulting equations will then have a solution for arbitrary αi if and only if a(s) and b(s) have no common factors.20

EXAMPLE 7.40

Pole Placement for Polynomial Transfer Functions Using the polynomial method, design a controller of order n for the third-order plant in Example 7.30. Note that if the polynomials αc (s) and αe (s) from Example 7.30 are multiplied, the result is the desired closed-loop characteristic equation: αc (s)αe (s) = s6 +14s5 +122.75s4 +585.2s3 +1505.64s2 +2476.8s+1728. (7.250) Solution. Using Eq. (7.249) with b(s) = 10, we find that (d0 s3 +d1 s2 +d2 s+d3 )(s3 +10s2 +16s)+10(c0 s2 +c1 s+c2 ) ≡ αc (s)αe (s). (7.251) We have expanded the polynomial d(s) with coefficients di and the polynomial cy (s) with coefficients ci . Now we equate the coefficients of the like powers of s in Eq. (7.251) to find that the parameters must satisfy21 ⎤⎡ ⎤ ⎡ ⎤ ⎡ d0 1 0 0 0 0 0 0 1 ⎢ ⎥ ⎢ 10 1 ⎥ ⎢ 0 0 0 0 0 ⎥ 14 ⎥ ⎢ d1 ⎥ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ 16 10 1 ⎥ 0 0 0 0 ⎥ ⎢ d2 ⎥ ⎢ 122.75 ⎥ ⎥ ⎢ ⎢ ⎥ ⎢ 0 ⎥ ⎢ 16 10 1 0 0 0 ⎥ (7.252) ⎥ ⎢ d3 ⎥ = ⎢ 585.2 ⎥ . ⎢ ⎢ ⎥ ⎢ 0 ⎢ ⎥ ⎥ 0 16 10 10 0 0 1505.64 c 0 ⎢ ⎥⎢ ⎥ ⎥ ⎢ ⎣ 0 0 0 16 0 10 0 ⎦ ⎣ c1 ⎦ ⎣ 2476.8 ⎦ 0 0 0 0 0 0 10 1728 c2 20 If they do have a common factor, it will show up on the left side of Eq. (7.249); for there to be a solution,

the same factor must be on the right side of Eq. (7.249), and thus a factor of either αc or αe . 21 The matrix on the left side of Eq. (7.252) is called a Sylvester matrix and is nonsingular if and only if a(s) and b(s) have no common factor.

526 Chapter 7 State-Space Design The solution to Eq. (7.252) is d0 = 1,

c0 = 190.1,

d1 = 4,

c1 = 481.8,

d2 = 66.75,

c2 = 172.8.

d3 = −146.3, MATLAB a\b

[The solution can be found using x = a\b command in MATLAB, where a is the Sylvester matrix and b is the right-hand side in Eq. (7.252).] Hence the controller transfer function is cy (s) 190.1s2 + 481.8s + 172.8 = 3 . d(s) s + 4s2 + 66.75s − 146.3

(7.253)

Note that the coefficients of Eq. (7.253) are the same as those of the controller Dc (s) (which we obtained using the state-variable techniques), once the factors in Dc (s) are multiplied out. The reduced-order compensator can also be derived using a polynomial solution.

EXAMPLE 7.41

Reduced-Order Design for a Polynomial Transfer Function Model Design a reduced-order controller for the third-order system in Example 7.30. The desired characteristic equation is αc (s)αe (s) = s5 + 12s4 + 74s3 + 207s2 + 378s + 288. Solution. The equations needed to solve this problem are the same as those used to obtain Eq. (7.251), except that we take both d(s) and cy (s) to be of degree n − 1. We need to solve (d0 s2 + d1 s + d2 )(s3 + 10s2 + 16s) + 10(c0 s2 + c1 s + c2 ) ≡ αc (s)αe (s). (7.254) Equating coefficients of like powers of s in Eq. (7.254), we obtain ⎡ ⎤ ⎡ ⎤⎡ ⎤ 1 0 0 0 0 0 1 d0 ⎢ 10 1 ⎢ ⎥ ⎢ ⎥ 0 0 0 0 ⎥ ⎢ ⎥ ⎢ d1 ⎥ ⎢ 12 ⎥ ⎢ 16 10 1 ⎥ ⎢ d2 ⎥ ⎢ 74 ⎥ 0 0 0 ⎢ ⎥=⎢ ⎥⎢ ⎥. ⎢ 0 ⎥ ⎢ ⎢ ⎥ 16 10 10 0 0 ⎥ ⎢ ⎥ ⎢ c0 ⎥ ⎢ 207 ⎥ ⎣ 0 0 16 0 10 0 ⎦ ⎣ c1 ⎦ ⎣ 378 ⎦ c2 0 0 0 0 0 10 288

MATLAB a\b

The solution is (again using the x = a\b command in MATLAB) d0 = 1,

c0 = −20.8,

d1 = 2.0,

c1 = −23.6,

d2 = 38,

c2 = 28.8,

(7.255)

7.13 Design for Systems with Pure Time Delay

527

and the resulting controller is cy (s) −20.8s2 − 23.6s + 28.8 = . d(s) s2 + 2.0s + 38

(7.256)

Again, Eq. (7.256) is exactly the same as Dcr (s) derived using the state-variable techniques in Example 7.31, once the polynomials of Dcr (s) are multiplied out and minor numerical differences are considered. Notice that the reference input polynomial cr (s) does not enter into the analysis of Examples 7.40 and 7.41. We can select cr (s) so that it will assign zeros in the transfer function from R(s) to Y (s). This is the same role played by γ (s) in Section 7.9. One choice is to select cr (s) to cancel αe (s) so that the overall transfer function is Y (s) Ks b(s) = . R(s) αc (s)

Adding integral control to the polynomial solution

This corresponds to the first and most common choice of M and N¯ for introducing the reference input described in Section 7.9. It is also possible to introduce integral control and, indeed, internal-model-based robust tracking control into the polynomial design method. What is required is that we have error control and that the controller have poles at the internal model locations. To get error control with the structure of Fig. 7.75, we need only let cr = cy . To get desired poles into the controller, we need to require that a specific factor be part of d(s). For integral control—the most common case—this is almost trivial. The polynomial d(s) will have a root at zero if we set the last term, dm , to zero. The resulting equations can be solved if m = n. For a more general internal model, we define d(s) to be the product of a reduced-degree polynomial and a specified polynomial such as Eq. (7.227), and match coefficients in the Diophantine equation as before. The process is straightforward but tedious. Again we caution that, while the polynomial design method can be effective, the numerical problems of this method are often much worse than those associated with methods based on state equations. For higher-order systems, the state-space methods are preferable.


7.13 Design for Systems with Pure Time Delay

Overall transfer function for a time-delayed system

In any linear system consisting of lumped elements, the response of the system appears immediately after an excitation of the system. In some feedback systems—for example, process control systems, whether controlled by a human operator in the loop or by computer—there is a pure time delay (also called transportation lag) in the system. As a result of the distributed nature of these systems, the response remains identically zero until after a delay of λ seconds. A typical step response is shown in Fig. 7.76(a). The transfer function of a pure transportation lag is e−λs . We can represent an overall transfer function of a single-input-single-output (SISO) system with time delay as GI (s) = G(s)e−λs , (7.257) where G(s) has no pure time delay. Because GI (s) does not have a finite state description, standard use of state-variable methods is impossible. However, Smith (1958)

528 Chapter 7 State-Space Design y (t)

1

t

td  l (a)

Compensator D(s) R






elsG(s)

D(s)

Y



R













elsG(s)

D(s)

Y

 G(s)  elsG(s)

(b)

(c)

Figure 7.76 A Smith regulator for systems with time delay showed how to construct a feedback structure that effectively takes the delay outside the loop and allows a feedback design based on G(s) alone, which can be done with standard methods. The result of this method is a design having closedloop transfer function with delay λ but otherwise showing the same response as the closed-loop design based on no delay. To see how the method works, let us consider the feedback structure shown in Fig. 7.76(b). The overall transfer function is Y (s) D
(s)G(s)e−λs = T (s) = . (7.258) R(s) 1 + D
(s)G(s)e−λs Smith suggested that we solve for D
(s) by setting up a dummy overall transfer function in which the controller transfer function D(s) is in a loop with G(s) with no loop delay but with an overall delay of λ: Y (s) D(s)G(s) −λs = T (s) = e . R(s) 1 + D(s)G(s) The Smith compensator

(7.259)

We then equate Eqs. (7.258) and (7.259) to solve for D
(s): D
(s) =

D(s) . 1 + D(s)[G(s) − G(s)e−λs ]

(7.260)

If the plant transfer function and the delay are known, D
(s) can be realized with real components by means of the block diagram shown in Fig. 7.76(c). With this knowledge, we can design the compensator D(s) in the usual way, based on

7.13 Design for Systems with Pure Time Delay

529

Eq. (7.259), as if there were no delay, and then implement it as shown in Fig. 7.76(c). The resulting closed-loop system would exhibit the behavior of a finite closed-loop system except for the time delay λ. This design approach is particularly suitable when the pure delay, λ, is significant as compared to the process time constant, for example, in pulp and paper process applications. Notice that, conceptually, the Smith compensator is feeding back a simulated plant output to cancel the true plant output and then adding in a simulated plant output without the delay. It can be demonstrated that D
(s) in Fig. 7.76(c) is equivalent to an ordinary regulator in line with a compensator that provides significant phase lead. To implement such compensators in analog systems, it is usually necessary to approximate the delay required in D
(s) by a Padé approximant; with digital compensators the delay can be implemented exactly (see Chapter 8). It is also a fact that the compensator D
(s) is a strong function of G(s), and a small error in the model of the plant used in the controller could lead to large errors in the closed loop, perhaps even to instability. This design is very sensitive. If D(s) is implemented as a PI controller, then one could detune (i.e., reduce the gain) to try to ensure stability and reasonable performance. For automatic tuning of the Smith regulator and a recent application to Stanford’s quiet hydraulic precision lathe fluid temperature control, the reader is referred to Huang and DeBra (2000).

EXAMPLE 7.42

Heat Exchanger: Design with Pure Time Delay Figure 7.77 shows the heat exchanger from Example 2.15. The temperature of the product is controlled by controlling the flow rate of steam in the exchanger jacket. The temperature sensor is several meters downstream from the steam control valve, which introduces a transportation lag into the model. A suitable model is given by G(s) =

Figure 7.77 A heat exchanger

e−5s . (10s + 1)(60s + 1)

Control valve Steam

Flow

Product Temperature sensor

Steam

530 Chapter 7 State-Space Design Design a controller for the heat exchanger using the Smith compensator and pole placement. The control poles are to be at pc = −0.05 ± 0.087j, and the estimator poles are to be at three times the control poles’ natural frequency: pe = −0.15 ± 0.26j. Simulate the response of the system with SIMULINK. Solution. A suitable set of state-space equations is     −0.017 0.017 0 x˙ (t) = x(t) + u(t − 5), 0 −0.1 0.1 y=[ 1

0 ]x,

λ = 5. For the specified control pole locations and for the moment ignoring the time delay, we find that the state feedback gain is K = [5.2 − 0.17]. For the given estimator poles, the estimator gain matrix for a full-order estimator is   0.18 L= . 4.2 The resulting controller transfer function is −0.25(s + 1.8) U(s) = . Y (s) s + 0.14 ± 0.27j If we choose to adjust for unity closed-loop DC gain, then N¯ = 1.2055. D(s) =

SIMULINK simulation

The SIMULINK diagram for the system is shown in Fig. 7.78. The open-loop and closed-loop step responses of the system and the control effort are shown in Figs. 7.79 and 7.80, and the root locus of the system (without the delay) is shown in Fig. 7.81. Note that the time delay of 5 sec in Figs. 7.79 and 7.80 is quite small compared with the response of the system, and is barely noticeable in this case.

7.14 Historical Perspective The state-variable approach to solving differential equations in engineering problems was advocated by R. E. Kalman while attending MIT. This was revolutionary and ruffled some feathers as it was going against the grain. The well-established academics, Kalman’s teachers, were well-versed in the frequency domain techniques and staunch supporters of it. Beginning in the late 1950s and early 1960s Kalman wrote a series of seminal papers introducing the ideas of state-variables, controllability, observability, the Linear Quadratic (LQ), and the Kalman Filter (LQF). Gunkel and Franklin (1963) and Joseph and Tou (1961) independently showed the separation theorem, which made possible the Linear Quadratic Gaussian (LQG) problem nowdays

531

7.14 Historical Perspective u To workspace1

1.20 Step

 

.25s  .45 s2  .28s  .0925 Transfer Fcn1

 

Gain

1 600s2  70s  1 Transfer Fcn2

Transport delay

Scope1

Scope

1 600s2  70s  1 Transfer Fcn

y To workspace

 

Transport delay

Figure 7.78 Closed-loop SIMULINK
diagram for a heat exchanger Figure 7.79 Step response for a heat exchanger

1.4 1.2 Output temperature, y

Closed loop 1 0.8 0.6 Open loop 0.4 0.2 0

0

50

100

150 200 Time (sec)

250

300

referred to as the H2 formulation. The separation theorem is a special case of the certainty-equivalence theorem of Simon (1956). The solutions to both LQ and LQG problems can be expressed in an elegant fashion in terms of the solutions to Riccati equations. D. G. Luenberger, who was taking a course with Kalman at Stanford University, derived the observer and reduced-order observer over a weekend after hearing Kalman suggesting the problem in a lecture. Kalman, Bryson, Athans, and others contributed to the field of optimal control theory that was widely employed in aerospace problems including the Apollo program. The book by Zadeh and Desoer published in 1962 was also influential in promoting the state-space method. In the

532 Chapter 7 State-Space Design Figure 7.80 Control effort for a heat exchanger

7 6

Control, u

5 4 3 2 1 0 1

0

50

100

Figure 7.81 Root locus for a heat exchanger

150 200 Time (sec)

250

300

Im(s) 0.3 0.2 Closed-loop poles

0.4 0.3 0.2 0.1

Closed-loop poles

0.1

0.2

0.3

0.4

Re(s)

0.2 0.3

1970s the robustness of LQ and LQG methods were studied resulting in the celebrated and influential paper of Doyle and Stein in 1981. One of the most significant contributions of Doyle and Safonov was to extend the idea of frequency domain gain to multi-input multi-output systems using the singular value decomposition (SVD). Others contributing to this research included G. Zames who introduced the H∞ methods that were found to be extensions to the H2 methods. The resulting design techniques are known as H∞ and µ-synthesis procedures. During the 1980s reliable numerical methods were developed for dealing with state-variable designs and computer-aided software for control design were developed. The invention of MATLAB by Cleve Moler and its wide distribution by The Mathworks has had a huge impact not only in the control design field but on all interactive scientific computations. While the state-variable methods were gaining momentum particularly in the US, research groups in Europe especially in England led by Rosenbrock, MacFarlane, Munro, and others extended the classical techniques to multi-input multi-output

Summary

533

systems. Hence root locus and frequency domain methods such as the (inverse) Nyquist techniques could be used for multi-input multi-output systems. Eventually in the 1980s there was a realization that the power of both frequency domain and state-variable methods should be combined for an eclectic control design method employing the best of both approaches. We saw in this Chapter 7 that in contrast to frequency response methods of Bode and Nyquist, the state-variable method not only deals with the input and output variables of the system but also with the internal physical variables. The state-variable methods can be used to study linear and nonlinear, as well as time varying systems. Furthermore, the state-variable method handles the multi-input multi-output problems and high-order systems with equal ease. From a computational perspective, the statevariable methods are far superior to the frequency domain techniques that require polynomial manipulations.

SUMMARY • • •

To every transfer function that has no more zeros than poles, there corresponds a differential equation in state-space form. State-space descriptions can be in several canonical forms. Among these are control, observer, and modal canonical forms. Open-loop poles and zeros can be computed from the state description matrices (F, G, H, J): Poles: p = eig(F), det(pI − F) = 0,   zI − F −G Zeros: det = 0. H J

•

• •

•

•

• •

For any controllable system of order n, there exists a state feedback control law that will place the closed-loop poles at the roots of an arbitrary control characteristic equation of order n. The reference input can be introduced so as to result in zero steady-state error to a step command. This property is not expected to be robust to parameter changes. Good closed-loop pole locations depend on the desired transient response, the robustness to parameter changes, and a balance between dynamic performance and control effort. Closed-loop pole locations can be selected to result in a dominant second-order response, to match a prototype dynamic response, or to minimize a quadratic performance measure. For any observable system of order n, an estimator (or observer) can be constructed with only sensor inputs and a state that estimates the plant state. The n poles of the estimator error system can be placed arbitrarily. Every transfer function can be represented by a minimal realization, i.e., a statespace model that is both controllable and observable. A single-input single-output system is completely controllable if and only if the input excites all the natural frequencies of the system, i.e., there is no cancellation of the poles in the transfer function.

534 Chapter 7 State-Space Design •

•

•

•

• •

•

• •

The control law and the estimator can be combined into a controller such that the poles of the closed-loop system are the sum of the control-law-only poles and the estimator-only poles. With the estimator-based controller, the reference input can be introduced in such a way as to permit n arbitrary zeros to be assigned. The most common choice is to assign the zeros to cancel the estimator poles, thus not exciting an estimator error. Integral control can be introduced to obtain robust steady-state tracking of a step by augmenting the plant state. The design is also robust with respect to rejecting constant disturbances. General robust control can be realized by combining the equations of the plant and the reference model into an error space and designing a control law for the extended system. Implementation of the robust design demonstrates the internal model principle. An estimator of the plant state can be added while retaining the robustness properties. The estimator can be extended to include estimates of the equivalent control disturbance and so result in robust tracking and disturbance rejection. Pole-placement designs, including integral control, can be computed using the polynomials of the plant transfer function in place of the state descriptions. Designs using polynomials frequently have problems with numerical accuracy. Controllers for plants that include a pure time delay can be designed as if there were no delay, and then a controller can be implemented for the plant with the delay. The design can be expected to be sensitive to parameter changes. Table 7.3 gives the important equations discussed in this chapter. The triangles indicate equations taken from optional sections in the text. Determining a model from experimental data, or verifying an analytically based model by experiment, is an important step in system design by state-space analysis, a step that is not necessarily needed for compensator design via frequency-response methods.

REVIEW QUESTIONS The following questions are based on a system in state-variable form with matrices F, G, H, J, input u, output y, and state x. 1. Why is it convenient to write equations of motion in state-variable form? 2.

Give an expression for the transfer function of this system.

3.

Give two expressions for the poles of the transfer function of the system.

4.

Give an expression for the zeros of the system transfer function.

5.

Under what condition will the state of the system be controllable?

6.

Under what conditions will the system be observable from the output y?

7.

Give an expression for the closed-loop poles if state feedback of the form u = −Kx is used.

8.

Under what conditions can the feedback matrix K be selected so that the roots of αc (s) are arbitrary?

9. What is the advantage of using the LQR or SRL in designing the feedback matrix K?

Review Questions

TABLE 7.3

535

Important Equations in Chapter 7 Name

Control canonical form

Equation ⎡

−a1 −a2 1 0 0 1 .. . 0 0 ⎡ ⎤ 1 ⎢0⎥ ⎢ ⎥ ⎢0⎥ ⎥ Bc = ⎢ ⎢ . ⎥, Cc = [b1 ⎢.⎥ ⎣.⎦

⎢ ⎢ ⎢ Ac = ⎢ ⎢ ⎢ ⎣

Page ··· ··· 0 .. . ···

b2

··· ··· ···

−an 0 0 .. . 0

0 1

···

···

⎤ ⎥ ⎥ ⎥ ⎥, ⎥ ⎥ ⎦

bn ],

426

Dc = 0.

0 State description Output equation Transformation of state

Controllability matrix Transfer function from state equations Transfer-function poles Transfer-function zeros Control characteristic equation Ackermann’s control formula for pole placement Reference input gains Control equation with reference input

Symmetric root locus Estimator error characteristic equation Observer canonical form

x˙ = Fx + Gu y = Hx + Ju A = T−1 FT B = T−1 G y = HTz + Ju = Cz + Du, where C = HT, D = J C = [G FG · · · Fn−1 G] Y (s) = H(sI − F)−1 G + J G(s) = U(s) det(pi I − F)= 0  z I − F −G =0 αz (s) = det i H J

428 428 429

det[sI − (F − GK)] = 0

444

K = [0 · · · 0 1]C −1 αc (F)      F G Nx 0 = H J Nu 1

448

u = Nu r − K(x − Nx r)

452

= −Kx + (Nu + KNx )r ¯ = −Kx + Nr 1 + ρG0 (−s)G0 (s) = 0

458

αe (s) = det[sI − (F − LH)] = 0

467

x˙ ◦ = F◦ x◦ + G◦ u, y = H◦ x◦ , where ⎡ −a1 1 0 0 ... ⎢ ⎢ 0 1 0 ... ⎢−a Fo = ⎢ 2 . . ⎢ .. .. .. ⎣ . 0 −an 0 ⎡ ⎤ b1 ⎢b2 ⎥ ⎢ ⎥ Go = ⎢ . ⎥ , Ho = [1 0 0 ⎣ .. ⎦ bn

470

430 437 438 439

452

⎤ 0 .⎥ .⎥ .⎥ ⎥, ⎥ 1⎦ 0

···

0],

536 Chapter 7 State-Space Design

TABLE 7.3

Continued Name

Controller equations

Equation ⎡ ⎤ H ⎢HF ⎥ ⎢ ⎥ ⎥ O=⎢ ⎢ .. ⎥ , ⎣ . ⎦ HFn−1 ⎡ ⎤ 0 ⎢0⎥ ⎥ ⎢ L = αe (F)O−1 ⎢ . ⎥ ⎣ .. ⎦ 1 U(s) Dc (s) = = −K(sI − F + GK + LH)−1 L Y (s) U(s) Dcr (s) = = Cr (sI − Ar )−1 Br + Dr Y (s) x˙ˆ = (F − GK − LH)ˆx + Ly + Mr

Augmented state equations with integral control

¯ u = −Kxˆ + Nr            0 1 x˙ I 0 H xI 0 r+ u− = + w G1 0 0 F G x˙ x

Observability matrix

Ackermann’s estimator formula Compensator transfer function Reduced-order compensator transfer function

 General controller in polynomial form  Diophantine equation for closed-loop characteristic equation

U(s) = −

cy (s) cr (s) Y (s) + R(s) d(s) d(s)

a(s)d(s) + b(s)cy (s) = αc (s)αe (s)

Page

471

471

480 480 492

503 524 524

10. What is the main reason for using an estimator in feedback control? 11. If the estimator gain L is used, give an expression for the closed-loop poles due to the estimator. 12.

Under what conditions can the estimator gain L be selected so that the roots of αe (s) = 0 are arbitrary?

13.

If the reference input is arranged so that the input to the estimator is identical to the input to the process, what will the overall closed-loop transfer function be?

14.

If the reference input is introduced in such a way as to permit the zeros to be assigned as the roots of γ (s), what will the overall closed-loop transfer function be?

15. What are the three standard techniques for introducing integral control in the state feedback design method?

PROBLEMS Problems for Section 7.3: Block Diagrams and State-Space 7.1

Write the dynamic equations describing the circuit in Fig. 7.82. Write the equations as a second-order differential equation in y(t). Assuming a zero input, solve the differential equation for y(t) using Laplace transform methods for the parameter values and

Problems L1H

Figure 7.82 Circuit for Problem 7.1

537

R2

C1F

u(t)

y(t)

y(to)  1 V, y(to)  0

initial conditions shown in the figure. Verify your answer using the initial command in MATLAB. 7.2

A schematic for the satellite and scientific probe for the Gravity Probe-B (GP-B) experiment that was launched on April 30, 2004 is sketched in Fig. 7.83. Assume that the mass of the spacecraft plus helium tank, m1 , is 2000 kg and the mass of the probe, m2 , is 1000 kg. A rotor will float inside the probe and will be forced to follow the probe with a capacitive forcing mechanism. The spring constant of the coupling k is 3.2 × 106 . The viscous damping b is 4.6 × 103 . (a) Write the equations of motion for the system consisting of masses m1 and m2 using the inertial position variables, y1 and y2 . (b) The actual disturbance u is a micrometeorite, and the resulting motion is very small. Therefore, rewrite your equations with the scaled variables z1 = 106 y1 , z2 = 106 y2 , and v = 1000u. (c) Put the equations in state-variable form using the state x = [z1 z˙1 z2 z˙2 ]T , the output y = z2 , and the input an impulse, u = 10−3 δ(t) N·sec on mass m1 . (d) Using the numerical values, enter the equations of motion into MATLAB in the form x˙ = Fx + Gv,

(7.261)

y = Hx + Jv.

(7.262)

and define the MATLAB system: sysGPB = ss(F,G,H,J). Plot the response of y caused by the impulse with the MATLAB command impulse(sysGPB). This is the signal the rotor must follow. (e) Use the MATLAB commands p = eig(F) to find the poles (or roots) of the system and z = tzero(F,G,H, J) to find the zeros of the system.

Figure 7.83 Schematic diagram of the GP-B satellite and probe

Rotor

k m1

u

m2 b

y1 y2

Problems for Section 7.4: Analysis of the State Equations 7.3

Give the state description matrices in control-canonical form for the following transfer functions: (a) 4s 1+ 1

538 Chapter 7 State-Space Design 5(s/2 + 1) (s/10 + 1) (c) 2 2s + 1 s + 3s + 2 s+3 (d) s(s2 + 2s + 2) + 10)(s2 + s + 25) (e) (s 2 s (s + 3)(s2 + s + 36)

(b)

7.4

Use the MATLAB function tf2ss to obtain the state matrices called for Problem 7.3.

7.5

Give the state description matrices in normal-mode form for the transfer functions of Problem 7.3. Make sure that all entries in the state matrices are real valued by keeping any pairs of complex conjugate poles together, and realize them as a separate subblock in control canonical form.

7.6

A certain system with state x is described by the state matrices     1 −2 1 , , G= F= −2 0 3  J = 0. H= 1 0 , Find the transformation T so that if x = Tz, the state matrices describing the dynamics of z are in control canonical form. Compute the new matrices A, B, C, and D.

7.7

Show that the transfer function is not changed by a linear transformation of state.

7.8

Use block-diagram reduction or Mason’s rule to find the transfer function for the system in observer canonical form depicted by Fig. 7.31.

7.9

Suppose we are given a system with state matrices F, G, H (J = 0 in this case). Find the transformation T so that, under Eqs. (7.24) and (7.25), the new state description matrices will be in observer canonical form.

7.10 Use the transformation matrix in Eq. (7.41) to explicitly multiply out the equations at the end of Example 7.10. 7.11 Find the state transformation that takes the observer canonical form of Eq. (7.35) to the modal canonical form. 7.12

(a) Find the transformation T that will keep the description of the tape-drive system of Example 7.11 in modal canonical form but will convert each element of the input matrix Bm to unity. (b) Use MATLAB to verify that your transformation does the job.

7.13

(a) Find the state transformation that will keep the description of the tape-drive system of Example 7.11 in modal canonical form but will cause the poles to be displayed in Am in order of increasing magnitude. (b) Use MATLAB to verify your result in part (a), and give the complete new set of state matrices as A, B, C, and D.

7.14 Find the characteristic equation for the modal-form matrix Am of Eq. (7.17a) using Eq. (7.58). 7.15 Given the system

 x˙ =

−4 −2

1 −1



 x+

0 1

 u

with zero initial conditions, find the steady-state value of x for a step input u.

Problems

539

7.16 Consider the system shown in Fig. 7.84: (a) Find the transfer function from U to Y . (b) Write state equations for the system using the state-variables indicated.

Figure 7.84 A block diagram for Problem 7.16

U






G1 

 x3

x4

1 s4

H2 

G2 



1 s




x2

H1 



1 2s

x1 




Y



1 s

7.17 Using the indicated state-variables, write the state equations for each of the systems shown in Fig. 7.85. Find the transfer function for each system using both block-diagram manipulation and matrix algebra [as in Eq. (7.48)].

Figure 7.85 Block diagrams for Problem 7.17

1 s4

U

x3



x2

s2 s3






x1



1 s



x1 




Y

5 (a)

1 s  10

U



x4 






1 s2

x3

4 s

x2

x1 1 s3 






Y

(b)

7.18 For each of the listed transfer functions, write the state equations in both control and observer canonical form. In each case, draw a block diagram and give the appropriate expressions for F, G, and H. (a)

s2 − 2 (control of an inverted pendulum by a force on the cart). 2 s (s2 − 1)

3s + 4 . s2 + 2s + 2 7.19 Consider the transfer function (b)

G(s) =

Y (s) s+1 . = 2 U(s) s + 5s + 6

(a) By rewriting Eq. (7.263) in the form G(s) =

1 s+3



(7.263)

 s+1 , s+2

find a series realization of G(s) as a cascade of two first-order systems. (b) Using a partial-fraction expansion of G(s), find a parallel realization of G(s). (c) Realize G(s) in control canonical form.

540 Chapter 7 State-Space Design Problems for Section 7.5: Control Law Design for Full-State Feedback 7.20 Consider the plant described by



x˙ =

0 7

y=[ 1

1 −4



 x+

1 2

 u,

3 ]x.

(a) Draw a block diagram for the plant with one integrator for each state-variable. (b) Find the transfer function using matrix algebra. (c) Find the closed-loop characteristic equation if the feedback is (i) u = −[ K1 K2 ]x; (ii)

u = −Ky.

7.21 For the system

   0 1 0 u, x+ 1 −6 −5 design a state feedback controller that satisfies the following specifications: 

x˙ =

(a) Closed-loop poles have a damping coefficient ζ = 0.707. (b) Step-response peak time is under 3.14 sec. Verify your design with MATLAB. 7.22

(a) Design a state feedback controller for the following system so that the closed-loop step response has an overshoot of less than 25% and a 1% settling time under 0.115 sec:     0 0 1 u. x+ x˙ = 1 0 −10 (b) Use the step command in MATLAB to verify that your design meets the specifications. If it does not, modify your feedback gains accordingly.

7.23 Consider the system

⎡

−1 x˙ = ⎣ 0 1

−2 −1 0

⎤ ⎤ ⎡ −2 2 1 ⎦ x + ⎣ 0 ⎦ u. −1 1

(a) Design a state feedback controller for the system so that the closed-loop step response has an overshoot of less than 5% and a 1% settling time under 4.6 sec. (b) Use the step command in MATLAB to verify that your design meets the specifications. If it does not, modify your feedback gains accordingly. 7.24 Consider the system in Fig. 7.86. (a) Write a set of equations that describes this system in the standard canonical control form as x˙ = Fx + Gu and y = Hx. (b) Design a control law of the form   x1 u = −[ K1 K2 ] x2 that will place the closed-loop poles at s = −2 ± 2j.

Figure 7.86 System for Problem 7.24

U

s s2  4

Y

Problems

541

7.25 Output Controllability. In many situations a control engineer may be interested in controlling the output y rather than the state x. A system is said to be output controllable if at any time you are able to transfer the output from zero to any desired output y∗ in a finite time using an appropriate control signal u∗ . Derive necessary and sufficient conditions for a continuous system (F, G, H) to be output controllable. Are output and state controllability related? If so, how? 7.26 Consider the system ⎡

0 ⎢ −1 ⎢ x˙ = ⎣ 5 0

4 −4 7 0

⎤ ⎡ 0 ⎢ 0 ⎥ ⎥x + ⎢ ⎣ 15 ⎦ −3

0 0 1 3

⎤ 0 0 ⎥ ⎥ u. 1 ⎦ 0

(a) Find the eigenvalues of this system. (Hint: Note the block-triangular structure.) (b) Find the controllable and uncontrollable modes of this system. (c) For each of the uncontrollable modes, find a vector v such that vT G = 0,

vT F = λvT .

(d) Show that there are an infinite number of feedback gains K that will relocate the modes of the system to −5, −3, −2, and −2. (e) Find the unique matrix K that achieves these pole locations and prevents initial conditions on the uncontrollable part of the system from ever affecting the controllable part. 7.27 Two pendulums, coupled by a spring, are to be controlled by two equal and opposite forces u, which are applied to the pendulum bobs as shown in Fig. 7.87. The equations of motion are ml2 θ¨1 = −ka2 (θ1 − θ2 ) − mglθ1 − lu, ml2 θ¨2 = −ka2 (θ2 − θ1 ) − mglθ2 + lu. (a) Show that the system is uncontrollable. Can you associate a physical meaning with the controllable and uncontrollable modes? (b) Is there any way that the system can be made controllable?

Figure 7.87 Coupled pendulums for Problem 7.27

a k l

u2

u1 u m

u m

542 Chapter 7 State-Space Design 7.28 The state-space model for a certain application has been given to us with the following state description matrices: ⎤ ⎡ ⎤ ⎡ −0.207 0.174 0 0 0 0 ⎢ −0.005 ⎥ ⎢ 0.157 0.645 0 0 0 ⎥ ⎥ ⎢ ⎥ ⎢ ⎥, G = ⎢ 0 ⎥, H = [ 1 0 0 0 0 ]. 0 1 0 0 0 F=⎢ ⎥ ⎢ ⎥ ⎢ ⎣ 0 ⎦ ⎣0 0 1 0 0⎦ 0 0 0 0 1 0 (a) Draw a block diagram of the realization with an integrator for each state-variable. (b) A student has computed det C = 2.3 × 10−7 and claims that the system is uncontrollable. Is the student right or wrong? Why? (c) Is the realization observable? 7.29 Staircase Algorithm (Van Dooren et al., 1978): Any realization (F, G, H) can be trans¯ G, ¯ H), ¯ where F¯ is an upper formed by an orthogonal similarity transformation to (F, Hessenberg matrix (having one nonzero diagonal above the main diagonal) given by ⎡

∗

⎢ ⎢ ¯F = TT FT = ⎢ ∗ ⎢ ⎣ ∗ ∗

α1 ∗ ∗ ∗

0 .. . .. . ···

0 0 αn−1 ∗

⎤ ⎥ ⎥ ⎥, ⎥ ⎦

⎡ ⎢ ¯ = TT G = ⎢ G ⎢ ⎣

0 .. . 0 g1

⎤ ⎥ ⎥ ⎥, ⎦

where g1 = 0, and ¯ = HT = [ h1 H

h2

···

hn ],

T−1 = TT .

Orthogonal transformations correspond to a rotation of the vectors (represented by the matrix columns) being transformed with no change in length. (a) Prove that if αi = 0 and αi+1 , . . . , αn−1 = 0 for some i, then the controllable and uncontrollable modes of the system can be identified after this transformation has been done. (b) How would you use this technique to identify the observable and unobservable modes of (F, G, H)? (c) What advantage does this approach for determining the controllable and uncontrollable modes have over transforming the system to any other form? (d) How can we use this approach to determine a basis for the controllable and uncontrollable subspaces, as in Problem 7.14? This algorithm can also be used to design a numerically stable algorithm for pole placement [see Minimis and Paige (1982)]. The name of the algorithm comes from the multi-input version in which the αi are the blocks that make F¯ resemble a staircase. Refer to ctrbf, obsvf commands in MATLAB.

Problems for Section 7.6: Selection of Pole Locations for Good Design 7.30 The normalized equations of motion for an inverted pendulum at angle θ on a cart are θ¨ = θ + u,

x¨ = −βθ − u,

where x is the cart position, and the control input u is a force acting on the cart.

Problems 

543

T

(a) With the state defined as x = θ θ˙ x x˙ , find the feedback gain K that places the closed-loop poles at s = −1, −1, −1 ± 1j. For parts (b) through (d), assume that β = 0.5. (b) Use the SRL to select poles with a bandwidth as close as possible to those of part (a), and find the control law that will place the closed-loop poles at the points you selected. (c) Compare the responses of the closed-loop systems in parts (a) and (b) to an initial condition of θ = 10◦ . You may wish to use the initial command in MATLAB. (d) Compute Nx and Nu for zero steady-state error to a constant command input on the cart position, and compare the step responses of each of the two closed-loop systems. 7.31 Consider the feedback system in Fig. 7.88. Find the relationship between K, T , and ξ such that the closed-loop transfer function minimizes the integral of the time multiplied by the absolute value of the error (ITAE) criterion,

∞ J = t|e| dt, 0

for a step input.

Figure 7.88 Control system for Problem 7.31

U








K 1  Ts

1 s2  2js  1

Y

7.32 Prove that the Nyquist plot for LQR design avoids a circle of radius one centered at the −1 point, as shown in Fig. 7.89. Show that this implies that 21 < GM < ∞, the “upward” gain margin is GM= ∞, and there is a “downward” GM = 21 , and the phase margin is at least PM = ±60◦ . Hence the LQR gain matrix, K, can be multiplied by a large scalar or reduced by half with guaranteed closed-loop system stability.

Figure 7.89 Nyquist plot for an optimal regulator

Im(L(jv))

a1 2

1

60

Re(L(jv))

Problems for Section 7.7: Estimator Design 7.33 Consider the system F=



−2 1

1 0



 ,G =

1 0

 ,H = [ 1

2 ],

544 Chapter 7 State-Space Design and assume that you are using feedback of the form u = −Kx + r, where r is a reference input signal. (a) Show that (F, H) is observable. (b) Show that there exists a K such that (F − GK, H) is unobservable. (c) Compute a K of the form K = [1, K2 ] that will make the system unobservable as in part (b); that is, find K2 so that the closed-loop system is not observable. (d) Compare the open-loop transfer function with the transfer function of the closedloop system of part (c). What is the unobservability due to? 7.34 Consider a system with the transfer function 9 . G(s) = 2 s −9 (a) (b) (c) (d) (e) (f)

Find (Fo , Go , Ho ) for this system in observer canonical form. Is (Fo , Go ) controllable? Compute K so that the closed-loop poles are assigned to s = −3 ± 3j. Is the closed-loop system of part (c) observable? Design a full-order estimator with estimator error poles at s = −12 ± 12j. Suppose the system is modified to have a zero: G1 (s) =

9(s + 1) . s2 − 9

Prove that if u = −Kx + r, there is a feedback gain K that makes the closed-loop system unobservable. [Again assume an observer canonical realization for G1 (s).] 7.35 Explain how the controllability, observability, and stability properties of a linear system are related. 7.36 Consider the electric circuit shown in Fig. 7.90. (a) Write the internal (state) equations for the circuit. The input u(t) is a current, and the output y is a voltage. Let x1 = iL and x2 = vc . (b) What condition(s) on R, L, and C will guarantee that the system is controllable? (c) What condition(s) on R, L, and C will guarantee that the system is observable?

Figure 7.90 Electric circuit for Problem 7.36

iL

L

 yc 

C

u R

R

 y 

7.37 The block diagram of a feedback system is shown in Fig. 7.91. The system state is  x=

xp xf

 ,

Problems Figure 7.91 Block diagram for Problem 7.37

r






G







xf



1 s




y

H

F

G

K

xp

1 s




545








L



 H

F

and the dimensions of the matrices are as follows: F = n × n,

L = n × 1,

G = n × 1,

x = 2n × 1,

H = 1 × n,

r = 1 × 1,

K = 1 × n,

y = 1 × 1.

(a) Write state equations for the system. (b) Let x = Tz, where  T=

I I

0 −I

 .

Show that the system is not controllable. (c) Find the transfer function of the system from r to y. 7.38 This problem is intended to give you more insight into controllability and observability. Consider the circuit in Fig. 7.92, with an input voltage source u(t) and an output current y(t). (a) Using the capacitor voltage and inductor current as state-variables, write state and output equations for the system. (b) Find the conditions relating R1 , R2 , C, and L that render the system uncontrollable. Find a similar set of conditions that result in an unobservable system. (c) Interpret the conditions found in part (b) physically in terms of the time constants of the system. (d) Find the transfer function of the system. Show that there is a pole–zero cancellation for the conditions derived in part (b) (that is, when the system is uncontrollable or unobservable).

Figure 7.92 Electric circuit for Problem 7.38

y(t)

 u(t) ⵑ 

C

 x1  R1

L

x2

R2

546 Chapter 7 State-Space Design 7.39 The linearized equations of motion for a satellite are x˙ = Fx + Gu, y = Hx, where ⎡ ⎢ F=⎢ ⎣  u=

⎤ 0 1 0 0 3ω2 0 0 2ω ⎥ ⎥, 0 0 0 1 ⎦ 0 −2ω 0 0    u1 y1 , y= . u2 y2

⎡

0 ⎢ 1 ⎢ G=⎣ 0 0

⎤ 0 0 ⎥ ⎥, 0 ⎦ 1

 H=

1 0

0 0

0 1

0 0

 ,

The inputs u1 and u2 are the radial and tangential thrusts, the state-variables x1 and x3 are the radial and angular deviations from the reference (circular) orbit, and the outputs y1 and y2 are the radial and angular measurements, respectively. (a) (b) (c) (d) 7.40

Show that the system is controllable using both control inputs. Show that the system is controllable using only a single input. Which one is it? Show that the system is observable using both measurements. Show that the system is observable using only one measurement. Which one is it?

Consider the system in Fig. 7.93. (a) Write the state-variable equations for the system, using [θ1 θ2 θ˙1 θ˙2 ]T as the state vector and F as the single input. (b) Show that all the state-variables are observable using measurements of θ1 alone. (c) Show that the characteristic polynomial for the system is the product of the polynomials for two oscillators. Do so by first writing a new set of system equations involving the state-variables ⎡ ⎤ ⎡ ⎤ θ1 + θ2 y1 ⎢ y2 ⎥ ⎢ θ1 − θ2 ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ y˙ 1 ⎦ = ⎣ θ˙1 + θ˙2 ⎦ . y˙ 2 θ˙1 − θ˙2

Figure 7.93 Coupled pendulums for Problem 7.40

d

k

u1

u2 F M

F M Gas jet K  kd u1  v2u1  K(u1  u2)  F/ml u2  v2u2  K(u1  u2)  F/ml

g

Problems Hint: If A and D are invertible matrices, then −1  −1  A A 0 = 0 D 0

547



0

.

D−1

(d) Deduce the fact that the spring mode is controllable with F but the pendulum mode is not. 7.41 A certain fifth-order system is found to have a characteristic equation with roots at 0, −1, −2, and −1 ± 1j. A decomposition into controllable and uncontrollable parts discloses that the controllable part has a characteristic equation with roots 0, and −1 ± 1j. A decomposition into observable and nonobservable parts discloses that the observable modes are at 0, −1, and −2. (a) Where are the zeros of b(s) = Hadj(sI − F)G for this system? (b) What are the poles of the reduced-order transfer function that includes only controllable and observable modes? 7.42 Consider the systems shown in Fig. 7.94, employing series, parallel, and feedback configurations. (a) Suppose we have controllable–observable realizations for each subsystem: x˙ i = Fi xi + Gi ui , yi = Hi xi ,

where i = 1, 2.

Give a set of state equations for the combined systems in Fig. 7.93. (b) For each case, determine what condition(s) on the roots of the polynomials Ni and Di is necessary for each system to be controllable and observable. Give a brief reason for your answer in terms of pole–zero cancellations. u1 u  u1

G1(s) 

N1(s) D1(s)

y1  u 2

G2(s) 

N2(s) D2(s)

y

G1(s) 

N1(s) D1(s)

G2(s) 

(a)

N2(s) D2(s) (b)






u1






u u2

r

y1

G1(s) 

N1(s) D1(s)

G2(s) 

N2(s) D2(s)



y

u2

(c)

Figure 7.94 Block diagrams for Problem 7.14: (a) series; (b) parallel; (c) feedback 7.43 Consider the system y¨ + 3˙y + 2y = u˙ + u.

y2



y  y1  y2

548 Chapter 7 State-Space Design (a) Find the state matrices Fc , Gc , and Hc in control canonical form that correspond to the given differential equation. (b) Sketch the eigenvectors of Fc in the (x1 , x2 ) plane, and draw vectors that correspond to the completely observable (x0 ) and the completely unobservable (x0¯ ) state-variables. (c) Express x0 and x0¯ in terms of the observability matrix O. (d) Give the state matrices in observer canonical form and repeat parts (b) and (c) in terms of controllability instead of observability. 7.44 The equations of motion for a station-keeping satellite (such as a weather satellite) are x¨ − 2ω˙y − 3ω2 x = 0,

y¨ + 2ω˙x = u,

where x = radial perturbation, y = longitudinal position perturbation, u = engine thrust in the y-direction, as depicted in Fig. 7.95. If the orbit is synchronous with the earth’s rotation, then ω = 2π/(3600 × 24) rad/sec.  T (a) Is the state x = x x˙ y y˙ observable? (b) Choose x = [ x x˙ y y˙ ]T as the state vector and y as the measurement, and design a full-order observer with poles placed at s = −2ω, −3ω, and −3ω ± 3ωj.

Figure 7.95 Diagram of a station-keeping satellite in orbit for Problem 7.44

u

Reference longitude

x y

Desired location on orbit

7.45 The linearized equations of motion of the simple pendulum in Fig. 7.96 are θ¨ + ω2 θ = u. (a) Write the equations of motion in state-space form. (b) Design an estimator (observer) that reconstructs the state of the pendulum given ˙ Assume ω = 5 rad/sec, and pick the estimator roots to be at measurements of θ. s = −10 ± 10j.

Problems

549

Figure 7.96 Pendulum diagram for Problem 7.45

u

(c) Write the transfer function of the estimator between the measured value of θ˙ and the estimated value of θ . (d) Design a controller (that is, determine the state feedback gain K) so that the roots of the closed-loop characteristic equation are at s = −4 ± 4j. 7.46 An error analysis of an inertial navigator leads to the set of normalized state equations ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ x1 x˙ 1 0 −1 0 0 ⎣ x˙ 2 ⎦ = ⎣ 1 0 1 ⎦ ⎣ x2 ⎦ + ⎣ 0 ⎦ u, 0 0 0 1 x˙ 3 x3 where x1 = east—velocity error, x2 = platform tilt about the north axis, x3 = north—gyro drift, u = gyro drift rate of change. Design a reduced-order estimator with y = x1 as the measurement, and place the observer-error poles at −0.1 and −0.1. Be sure to provide all the relevant estimator equations.

Problems for Section 7.8: Compensator Design: Combined Control Law and Estimator 7.47 A certain process has the transfer function G(s) =

4 . (s2 −4)

(a) Find F, G, and H for this system in observer canonical form. (b) If u = −Kx, compute K so that the closed-loop control poles are located at s = −2 ± 2j. (c) Compute L so that the estimator error poles are located at s = −10 ± 10j. (d) Give the transfer function of the resulting controller [for example, using Eq. (7.177)]. (e) What are the gain and phase margins of the controller and the given open-loop system? 7.48 The linearized longitudinal motion of a helicopter near hover (see Fig. 7.97) can be modeled by the normalized third-order system ⎤ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎡ q˙ 6.3 q −0.4 0 −0.01 ⎦ ⎣ θ ⎦ + ⎣ 0 ⎦ δ, ⎣ θ˙ ⎦ = ⎣ 1 0 0 9.8 u −1.4 9.8 −0.02 u˙

550 Chapter 7 State-Space Design Figure 7.97 Helicopter for Problem 7.48

Fuselage reference axis Vertical

u

d Rotor thrust

Rotor u

q = pitch rate, θ = pitch angle of fuselage, u = horizontal velocity (standard aircraft notation), δ = rotor tilt angle (control variable). Suppose our sensor measures the horizontal velocity u as the output; that is, y = u. (a) Find the open-loop pole locations. (b) Is the system controllable? (c) Find the feedback gain that places the poles of the system at s = −1 ± 1j and s = −2. (d) Design a full-order estimator for the system, and place the estimator poles at −8 √ and −4 ± 4 3j. (e) Design a reduced-order estimator with both poles at −4. What are the advantages and disadvantages of the reduced-order estimator compared with the full-order case? (f) Compute the compensator transfer function using the control gain and the full-order estimator designed in part (d), and plot its frequency response using MATLAB. Draw a Bode plot for the closed-loop design, and indicate the corresponding gain and phase margins. (g) Repeat part (f) with the reduced-order estimator. (h) Draw the SRL and select roots for a control law that will give a control bandwidth matching the design of part (c), and select roots for a full-order estimator that will result in an estimator error bandwidth comparable to the design of part (d). Draw the corresponding Bode plot and compare the pole placement and SRL designs with respect to bandwidth, stability margins, step response, and control effort for a unit-step rotor-angle input. Use MATLAB for the computations. 7.49 Suppose a DC drive motor with motor current u is connected to the wheels of a cart in order to control the movement of an inverted pendulum mounted on the cart. The linearized and normalized equations of motion corresponding to this system can be put in the form θ¨ = θ + v + u, v˙ = θ − v − u,

Problems

551

where θ =angle of the pendulum, v = velocity of the cart. (a) We wish to control θ by feedback to u of the form u = −K1 θ − K2 θ˙ − K3 v. Find the √feedback gains so that the resulting closed-loop poles are located at −1, −1 ± j 3. (b) Assume that θ and v are measured. Construct an estimator for θ and θ˙ of the form x˙ˆ = Fˆx + L(y − yˆ ), ˙ T and y = θ . Treat both v and u as known. Select L so that the where x = [θ θ] estimator poles are at −2 and −2. (c) Give the transfer function of the controller, and draw the Bode plot of the closed-loop system, indicating the corresponding gain and phase margins. (d) Using MATLAB, plot the response of the system to an initial condition on θ, and give a physical explanation for the initial motion of the cart. 7.50 Consider the control of

Y (s) 10 = . U(s) s(s + 1) Let y = x1 and x˙ 1 = x2 , and write state equations for the system. Find K1 and K2 so that u = −K1 x1 − K2 x2 yields closed-loop poles with a natural frequency ωn = 3 and a damping ratio ζ = 0.5. Design a state estimator for the system that yields estimator error poles with ωn1 = 15 and ζ1 = 0.5. What is the transfer function of the controller obtained by combining parts (a) through (c)? Sketch the root locus of the resulting closed-loop system as plant gain (nominally 10) is varied. G(s) =

(a) (b) (c) (d) (e) 7.51

Unstable equations of motion of the form x¨ = x + u arise in situations where the motion of an upside-down pendulum (such as a rocket) must be controlled. (a) Let u = −Kx (position feedback alone), and sketch the root locus with respect to the scalar gain K. (b) Consider a lead compensator of the form U(s) = K

s+a X(s). s + 10

Select a and K so that the system will display a rise time of about 2 sec and no more than 25% overshoot. Sketch the root locus with respect to K. (c) Sketch the Bode plot (both magnitude and phase) of the uncompensated plant. (d) Sketch the Bode plot of the compensated design, and estimate the phase margin.

552 Chapter 7 State-Space Design (e) Design state feedback so that the closed-loop poles are at the same locations as those of the design in part (b). (f) Design an estimator for x and x˙ using the measurement of x = y, and select the observer gain L so that the equation for x˜ has characteristic roots with a damping ratio ζ = 0.5 and a natural frequency ωn = 8. (g) Draw a block diagram of your combined estimator and control law, and indicate where xˆ and x˙ appear. Draw a Bode plot for the closed-loop system, and compare the resulting bandwidth and stability margins with those obtained using the design of part (b). 7.52 A simplified model for the control of a flexible robotic arm is shown in Fig. 7.98, where k/M = 900 rad/sec2 , y = output, the mass position, u = input, the position of the end of the spring. (a) Write the equations of motion in state-space form. (b) Design an estimator with roots as s = −100 ± 100j. (c) Could both state-variables of the system be estimated if only a measurement of y˙ was available? (d) Design a full-state feedback controller with roots at s = −20 ± 20j. (e) Would it be reasonable to design a control law for the system with roots at s = −200 ± 200j? State your reasons. (f) Write equations for the compensator, including a command input for y. Draw a Bode plot for the closed-loop system and give the gain and phase margins for the design.

Figure 7.98 Simple robotic arm for Problem 7.52

u

y k M

7.53 The linearized differential equations governing the fluid-flow dynamics for the two cascaded tanks in Fig. 7.99 are δ h˙ 1 + σ δh1 = δu, δ h˙ 2 + σ δh2 = σ δh1 , where δh1 = deviation of depth in tank 1 from the nominal level, δh2 = deviation of depth in tank 2 from the nominal level, δu = deviation in fluid inflow rate to tank 1 (control). (a) Level Controller for Two Cascaded Tanks: Using state feedback of the form δu = −K1 δh1 − K2 δh2 ,

Problems Figure 7.99 Coupled tanks for Problem 7.53

553

u

h1

h2

choose values of K1 and K2 that will place the closed-loop eigenvalues at s = −2σ (1 ± j). (b) Level Estimator for Two Cascaded Tanks: Suppose that only the deviation in the level of tank 2 is measured (that is, y = δh2 ). Using this measurement, design an estimator that will give continuous, smooth estimates of the deviation in levels of tank 1 and tank 2, with estimator error poles at −8σ (1 ± j). (c) Estimator/Controller for Two Cascaded Tanks: Sketch a block diagram (showing individual integrators) of the closed-loop system obtained by combining the estimator of part (b) with the controller of part (a). (d) Using MATLAB, compute and plot the response at y to an initial offset in δh1 . Assume σ = 1 for the plot. 7.54 The lateral motions of a ship that is 100 10 m/sec, are described by ⎡ ⎤ ⎡ β˙ −0.0895 −0.286 ⎣ r˙ ⎦ = ⎣ −0.0439 −0.272 0 1 ψ˙

m long, moving at a constant velocity of ⎤ ⎡ ⎤ ⎤⎡ β 0.0145 0 0 ⎦ ⎣ r ⎦ + ⎣ −0.0122 ⎦ δ, ψ 0 0

where β = sideslip angle (deg), ψ = heading angle (deg), δ = rudder angle (deg), r = yaw rate (see Fig. 7.99). (a) Determine the transfer function from δ to ψ and the characteristic roots of the uncontrolled ship. (b) Using complete state feedback of the form δ = −K1 β − K2 r − K3 (ψ − ψd ), where ψd is the desired heading, determine values of K1 , K2 , and K3 that will place the closed-loop roots at s = −0.2, −0.2 ± 0.2j. (c) Design a state estimator based on the measurement of ψ (obtained from a gyrocompass, for example). Place the roots of the estimator error equation at s = −0.8 and −0.8 ± 0.8j.

554 Chapter 7 State-Space Design (d) Give the state equations and transfer function for the compensator Dc (s) in Fig. 7.100, and plot its frequency response. (e) Draw the Bode plot for the closed-loop system, and compute the corresponding gain and phase margins. (f) Compute the feed-forward gains for a reference input, and plot the step response of the system to a change in heading of 5◦ .

Figure 7.100 View of ship from above for Problem 7.54

b

Ship motion

c

d

Figure 7.101 Ship control block diagram for Problem 7.54

d

G

c

Dc cd

Problems for Section 7.9: Introduction of the Reference Input with the Estimator 

7.55 As mentioned in footnote 11 in Section 7.9.2, a reasonable approach for selecting the feed-forward gain in Eq. (7.205) is to choose N¯ such that when r and y are both unchanging, the DC gain from r to u is the negative of the DC gain from y to u. Derive a formula for N¯ based on this selection rule. Show that if the plant is Type 1, this choice is the same as that given by Eq. (7.205).

Problems

555

Problems for Section 7.10: Integral Control and Robust Tracking 7.56 Assume that the linearized and time-scaled equation of motion for the ball-bearing levitation device is x¨ − x = u + w. Here w is a constant bias due to the power amplifier. Introduce integral error control, and select three control gains K = [ K1 K2 K3 ] so that the closed-loop poles are at −1 and −1 ± j and the steady-state error to w and to

a (step) position command will be zero. Let y = x and the reference input r = yref be a constant. Draw a block diagram of your design showing the locations of the feedback gains Ki . Assume that both x˙ and x can be measured. Plot the response of the closed-loop system to a step command input and the response to a step change in the bias input. Verify that the system is Type 1. Use MATLAB (SIMULINK) software to simulate the system responses. 7.57

Consider a system with state matrices     1 −2 1 , , G= F= 1 0 −3

H=[ 1

3 ].

¯ (a) Use feedback of the form u(t) = −Kx(t) + Nr(t), where N¯ is a nonzero scalar, to move the poles to −3 ± 3j. (b) Choose N¯ so that if r is a constant, the system has zero steady-state error; that is y(∞) = r. (c) Show that if F changes to F + δF, where δF is an arbitrary 2 × 2 matrix, then your choice of N¯ in part (b) will no longer make y(∞) = r. Therefore, the system is not robust under changes to the system parameters in F. (d) The system steady-state error performance can be made robust by augmenting the system with an integrator and using unity feedback—that is, by setting x˙ I = r − y, where xI is the state of the integrator. To see this, first use state feedback of the form √ u = −Kx − K1 xI so that the poles of the augmented system are at −3, −2 ± j 3. (e) Show that the resulting system will yield y(∞) = r no matter how the matrices F and G are changed, as long as the closed-loop system remains stable. (f) For part (d), use MATLAB (SIMULINK) software to plot the time response of the system to a constant input. Draw Bode plots of the controller, as well as the sensitivity function (S) and the complementary sensitivity function (T ).



7.58 Consider a servomechanism for following the data track on a computer-disk memory system. Because of various unavoidable mechanical imperfections, the data track is not exactly a centered circle, and thus the radial servo must follow a sinusoidal input of radian frequency ω0 (the spin rate of the disk). The state matrices for a linearized model of such a system are     0 0 1 , H = [ 1 3 ]. , G= F= 1 0 −1 The sinusoidal reference input satisfies r¨ = −ω02 r. (a) Let ω0 = 1, and place the poles of the error system for an internal model design at αc (s) = (s + 2 ± j2)(s + 1 ± 1j) and the pole of the reduced-order estimator at αe (s) = (s + 6).

556 Chapter 7 State-Space Design (b) Draw a block diagram of the system, and clearly show the presence of the oscillator with frequency ω0 (the internal model) in the controller. Also verify the presence of the blocking zeros at ±jω0 . (c) Use MATLAB (SIMULINK) software to plot the time response of the system to a sinusoidal input at frequency ω0 = 1. (d) Draw a Bode plot to show how this system will respond to sinusoidal inputs at frequencies different from but near ω0 .



7.59 Compute the controller transfer function [from Y (s) to U(s)] in Example 7.38. What is the prominent feature of the controller that allows tracking and disturbance rejection?



7.60 Consider the pendulum problem with control torque Tc and disturbance torque Td : θ¨ + 4θ = Tc + Td . (Here g/l = 4.) Assume that there is a potentiometer at the pin that measures the output angle θ, but with a constant unknown bias b. Thus the measurement equation is y = θ + b. (a) Take the “augmented” state vector to be ⎡

⎤ θ ⎣ θ˙ ⎦ , w

where w is the input-equivalent bias. Write the system equations in state-space form. Give values for the matrices F, G, and H. (b) Using state-variable methods, show that the characteristic equation of the model is s(s2 + 4) = 0. (c) Show that w is observable if we assume that y = θ , and write the estimator equations for ⎡ ⎤ θˆ ⎢ ˙ ⎥ ⎣ θˆ ⎦ . wˆ Pick estimator gains [ l1 l2 l3 ]T to place all the roots of the estimator error characteristic equation at −10. (d) Using full-state feedback of the estimated (controllable) state-variables, derive a control law to place the closed-loop poles at −2 ± 2j. (e) Draw a block diagram of the complete closed-loop system (estimator, plant, and controller) using integrator blocks. (f) Introduce the estimated bias into the control so as to yield zero steady-state error to the output bias b. Demonstrate the performance of your design by plotting the response of the system to a step change in b; that is, b changes from 0 to some constant value.

Problems for Section 7.13: Design for Systems with Pure Time Delay 

7.61 Consider the system with the transfer function e−Ts G(s), where G(s) =

1 . s(s + 1)(s + 2)

Problems

557

The Smith compensator for this system is given by Dc
(s) =

Dc . 1 + (1 − e−sT )G(s)Dc

Plot the frequency response of the compensator for T = 5 and Dc = 1, and draw a Bode plot that shows the gain and phase margins of the system.22

22 This problem was given by Åström (1977).

8

Digital Control

A Perspective on Digital Control Most of the controllers we have studied so far were described by the Laplace transform or differential equations, which, strictly speaking, are assumed to be built using analog electronics, such as those in Figs. 5.31 and 5.35. However, as discussed in Section 4.4, most control systems today use digital computers (usually microprocessors) to implement the controllers. The intent of this chapter is to expand on the design of control systems that will be implemented in a digital computer. The implementation leads to an average delay of half the sample period and to a phenomenon called aliasing, which need to be addressed in the controller design. Analog electronics can integrate and differentiate signals. In order for a digital computer to accomplish these tasks, the differential equations describing compensation must be approximated by reducing them to algebraic equations involving addition, division, and multiplication, as developed in Section 4.4. This chapter expands on various ways to make these approximations. The resulting design can then be tuned up, if needed, using direct digital analysis and design. You should be able to design, analyze, and implement a digital control system from the material in this chapter. However, our treatment here is a

558

8.1 Digitization

559

limited version of a complex subject covered in more detail in Digital Control of Dynamic Systems by Franklin et al. (1998 3rd ed.).

Chapter Overview In Section 8.1 we describe the basic structure of digital control systems and introduce the issues that arise due to the sampling. The digital implementation described in Section 4.4 is sufficient for implementing a feedback control law in a digital control system, which you can then evaluate via SIMULINK
to determine the degradation with respect to the continuous case. However, to fully understand the effect of sampling, it is useful to learn about discrete linear analysis tools. This requires an understanding of the z-transform, which we discuss in Section 8.2. Section 8.3 builds on this understanding to provide a better foundation for design using discrete equivalents that was briefly discussed in Section 4.4. Hardware characteristics and sample rate issues are discussed in Sections 8.4 and 8.5, both of which need to be addressed in order to implement a digital controller. In contrast to discrete equivalent design, which is an approximate method, optional Section 8.6 explores direct digital design (also called discrete design), which provides an exact method that is independent of whether the sample rate is fast or not.

8.1 Digitization

Sample period

Figure 8.1(a) shows the topology of the typical continuous system that we have been considering in previous chapters. The computation of the error signal e and the dynamic compensation D(s) can all be accomplished in a digital computer as shown in Fig. 8.1(b). The fundamental differences between the two implementations are that the digital system operates on samples of the sensed plant output rather than on the continuous signal and that the control provided by D(s) must be generated by algebraic recursive equations. We consider first the action of the analog-to-digital (A/D) converter on a signal. This device samples a physical variable, most commonly an electrical voltage, and converts it into a binary number that usually consists of 10 to 16 bits. Conversion from the analog signal y(t) to the samples, y(kT ), occurs repeatedly at instants of time T seconds apart. T is the sample period, and 1/T is the sample rate in Hertz. The sampled signal is y(kT ), where k can take on any integer value. It is often written simply as y(k). We call this type of variable a discrete signal to distinguish it from a continuous signal such as y(t), which changes continuously in time. A system having both discrete and continuous signals is called a sampled data system. We make the assumption that the sample period is fixed. In practice, digital control systems sometimes have varying sample periods and/or different periods in different feedback paths. Usually, the computer logic includes a clock that supplies a pulse, or interrupt, every T seconds, and the A/D converter sends a number to the computer each time the interrupt arrives. An alternative implementation, often referred to as free running, is to access the A/D converter after each cycle of code execution has been completed. In the former case the sample period is precisely fixed; in the latter

560 Chapter 8 Digital Control Figure 8.1 Block diagrams for a basic control system: (a) continuous system; (b) with a digital computer

Continuous controller 

r(t)

e(t)




D(s)

u(t)

Plant y(t)

G(s)

 Sensor y(t) 1 (a)

Digital controller 

r(t) T

r(kT)




e(kT)



Plant

Difference u(kT) D/A and equations hold

u(t)

G(s)

y(t)

Clock y(kT)

Sampler T y(t)

A/D

Sensor 1

(b)

Zero-order hold (ZOH)

Discrete equivalents

Sample rate selection

case the sample period is fixed essentially by the length of the code, provided that no logic branches are present, which could vary the amount of code executed. There also may be a sampler and an A/D converter for the input command r(t), which produces the discrete r(kT ), from which the sensed output y(kT ) will be subtracted to arrive at the discrete error signal e(kT ). As we saw in Sections 4.4 and 5.4.4, and Example 6.15, the continuous compensation is approximated by difference equations, which are the discrete version of differential equations and can be made to duplicate the dynamic behavior of D(s) if the sample period is short enough. The result of the difference equations is a discrete u(kT ) at each sample instant. This signal is converted to a continuous u(t) by the digital-to-analog (D/A) converter and the hold: the D/A converter changes the binary number to an analog voltage, and a zero-order hold maintains that same voltage throughout the sample period. The resulting u(t) is then applied to the actuator in precisely the same manner as the continuous implementation. There are two basic techniques for finding the difference equations for the digital controller. One technique, called discrete equivalent, consists of designing a continuous compensation D(s) using methods described in the previous chapters, then approximating that D(s) using the method of Section 4.4 (Tustin’s Method), or one of the other methods described in Section 8.3. The other technique is discrete design, described in Section 8.6. Here the difference equations are found directly without designing D(s) first. The sample rate required depends on the closed-loop bandwidth of the system. Generally, sample rates should be about 20 times the bandwidth or faster in order to assure that the digital controller will match the performance of the continuous

8.2 Dynamic Analysis of Discrete Systems Figure 8.2 The delay due to the hold operation

561

u u(kT) Continuous control, u(t)

Average u(t) Control from D/A

1

2

3

4

5

6

7

8

kT

controller. Slower sample rates can be used if some adjustments are made in the digital controller or some performance degradation is acceptable. Use of the discrete design method described in Section 8.6 allows for a much slower sample rate if that is desirable to minimize hardware costs; however, best performance of a digital controller is obtained when the sample rate is greater than 25 times the bandwidth. It is worth noting that the single most important impact of implementing a control system digitally is the delay associated with the hold. Because each value of u(kT ) in Fig. 8.1(b) is held constant until the next value is available from the computer, the continuous value of u(t) consists of steps (see Fig. 8.2) that, on average, are delayed from u(kT ) by T/2 as shown in the figure. If we simply incorporate this T/2 delay into a continuous analysis of the system, an excellent prediction of the effects of sampling results for sample rates much slower than 20 times bandwidth. We will discuss this further in Section 8.3.3.

8.2 Dynamic Analysis of Discrete Systems The z-transform is the mathematical tool for the analysis of linear discrete systems. It plays the same role for discrete systems that the Laplace transform does for continuous systems. This section will give a short description of the z-transform, describe its use in analyzing discrete systems, and show how it relates to the Laplace transform.

8.2.1 z-Transform In the analysis of continuous systems, we use the Laplace transform, which is defined by

∞ L{f (t)} = F(s) = f (t)e−st dt, 0

which leads directly to the important property that (with zero initial conditions) L{f˙ (t)} = sF(s).

(8.1)

Relation (8.1) enables us easily to find the transfer function of a linear continuous system, given the differential equation of that system.

562 Chapter 8 Digital Control Figure 8.3 A continuous, sampled version of signal f

f(t) f(k) f(t)

T 0

z-transform

T

2T

3T

t

For discrete systems a similar procedure is available. The z-transform is defined by ∞  Z{f (k)} = F(z) = f (k)z−k , (8.2) k=0

where f (k) is the sampled version of f (t), as shown in Fig. 8.3, and k = 0, 1, 2, 3, . . . refers to discrete sample times t0 , t1 , t2 , t3 , . . . . This leads directly to a property analogous to Eq. (8.1), specifically, that Z{f (k − 1)} = z−1 F(z).

(8.3)

This relation allows us to easily find the transfer function of a discrete system, given the difference equations of that system. For example, the general second-order difference equation y(k) = −a1 y(k − 1) − a2 y(k − 2) + b0 u(k) + b1 u(k − 1) + b2 u(k − 2) can be converted from this form to the z-transform of the variables y(k), u(k), . . . by invoking Eq. (8.3) once or twice to arrive at Y (z) = (−a1 z−1 − a2 z−2 )Y (z) + (b0 + b1 z−1 + b2 z−2 )U(z). Discrete transfer function

(8.4)

Equation (8.4) then results in the discrete transfer function b0 + b1 z−1 + b2 z−2 Y (z) = . U(z) 1 + a1 z−1 + a2 z−2

8.2.2 z-Transform Inversion Table 8.1 relates simple discrete-time functions to their z-transforms and gives the Laplace transforms for the same time functions. Given a general z-transform, we could expand it into a sum of elementary terms using partial-fraction expansion (see Appendix A) and find the resulting time series from the table. These procedures are exactly the same as those used for continuous systems; as with the continuous case, most designers would use a numerical evaluation of the discrete equations to obtain a time history rather than inverting the z-transform. A z-transform inversion technique that has no continuous counterpart is called long division. Given the z-transform Y (z) =

N(z) , D(z)

(8.5)

8.2 Dynamic Analysis of Discrete Systems

TABLE 8.1

563

Laplace Transforms and z-Transforms of Simple Discrete-Time Functions No. 1 2 3

F (s)

f (kT )

F (z) 1 z−ko

1 s

1, k = 0; 0, k = 0 1, k = ko ; 0, k = ko 1(kT )

z z−1

4

1 s2

kT

Tz (z−1)2

5

1 s3

1 2 2! (kT )

T2 2

6

1 s4

1 3 3! (kT )

T3 6

7

1 sm

m−1 lima→0 (−1) (m−1)!

8

1 s+a

e−akT

z z−e−aT

9

1 (s+a)2

kTe−akT

Tze−aT (z−e−aT )2

10

1 (s+a)3

1 2 −akT 2 (kT ) e

T 2 e−aT z (z+e−aT ) 2 (z−e−aT )3

11

1 (s+a)m

(−1)m−1 (m−1)!

12

a s(s+a)

1 − e−akT

z(1−e−aT ) (z−1)(z−e−aT )

13

a s2 (s+a)

1 −akT ) a (akT − 1 + e

z[(aT −1+e−aT )z+(1−e−aT −aTe−aT )] a(z−1)2 (z−e−aT )

14

b−a (s+a)(s+b)

e−akT − e−bkT

(e−aT −e−bT )z (z−e−aT )(z−e−bT )

15

s (s+a)2

(1 − akT )e−akT

z[z−e−aT (1+aT )] (z−e−aT )2

16

a2 s(s+a)2

1 − e−akT (1 + akT )

z[z(1−e−aT −aTe−aT )+e−2aT −e−aT +aTe−aT ] (z−1)(z−e−aT )2

17

(b−a)s (s+a)(s+b)

be−bkT − ae−akT

z[z(b−a)−(be−aT −ae−bT )] (z−e−aT )(z−e−bT )

18

a s2 +a2

sin akT

z sin aT z2 −(2 cos aT )z+1

19

s s2 +a2

cos akT

z(z−cos aT ) z2 −(2 cos aT )z+1

20

s+a (s+a)2 +b2

e−akT cos bkT

z(z−e−aT cos bT ) z2 −2e−aT (cos bT )z+e−2aT

21

b (s+a)2 +b2

e−akT sin bkT

ze−aT sin bT z2 −2e−aT (cos bT )z+e−2aT

22

a2 +b2 s[(s+a)2 +b2 ]

1 − e−akT (cos bkT + ab sin bkT )

z(Az+B) (z−1)[z2 −2e−aT (cos bT )z+e−2aT ]







∂ m−1 e−akT ∂am−1

∂ m−1 e−akT ∂am−1









z(z+1) (z−1)3



z(z2 +4z+1) (z−1)4



m−1 lima→0 (−1) (m−1)!

(−1)m−1 (m−1)!







∂ m−1 z ∂am−1 z−e−aT

∂ m−1 z ∂am−1 z−e−aT





A = 1 − e−aT cos bT − ab e−aT sin bT B = e−2aT + ab e−aT sin bT − e−aT cos bT F(s) is the Laplace transform of f (t), and F (z) is the z -transform of f (kT ). Note: f (t) = 0 for t = 0.

564 Chapter 8 Digital Control z-transform inversion: long division

we simply divide the denominator into the numerator using long division. The result is a series (perhaps with an infinite number of terms) in z−1 , from which the time series can be found by using Eq. (8.2). For example, a first-order system described by the difference equation y(k) = αy(k − 1) + u(k) yields the discrete transfer function 1 Y (z) = . U(z) 1 − αz−1 For a unit-pulse input defined by u(0) = 1, u(k) = 0

k = 0,

the z-transform is then U(z) = 1,

(8.6)

so

1 . (8.7) 1 − αz−1 Therefore, to find the time series, we divide the numerator of Eq. (8.7) by its denominator using long division: 1 + αz−1 + α 2 z−2 + α 3 z−3 + · · · Y (z) =

1 − αz−1

+

1 1 − αz−1 αz−1 + 0 αz−1 − α 2 z−2 α 2 z−2 + 0 α 2 z−2 − α3z−3 α 3 z−3 .. .

This yields the infinite series Y (z) = 1 + αz−1 + α 2 z−2 + α 3 z−3 + · · · . From Eqs. (8.8) and (8.2) we see that the sampled time history of y is y(0) = 1, y(1) = α, y(2) = α 2 , .. .

.. .

y(k) = α k .

(8.8)

8.2 Dynamic Analysis of Discrete Systems

565

8.2.3 Relationship between s and z For continuous systems, we saw in Chapter 3 that certain behaviors result from different pole locations in the s-plane: oscillatory behavior for poles near the imaginary axis, exponential decay for poles on the negative real axis, and unstable behavior for poles with a positive real part. A similar kind of association would also be useful to know when designing discrete systems. Consider the continuous signal f (t) = e−at ,

t > 0,

which has the Laplace transform F(s) =

1 s+a

and corresponds to a pole at s = −a. The z-transform of f (kT ) is F(z) = Z{e−akT }.

(8.9)

From Table 8.1 we can see that Eq. (8.9) is equivalent to F(z) =

z , z − e−aT

which corresponds to a pole at z = e−aT . This means that a pole at s = −a in the s-plane corresponds to a pole at z = e−aT in the discrete domain. This is true in general: Relationship between z -plane and s -plane characteristics

The equivalent characteristics in the z-plane are related to those in the s-plane by the expression (8.10) z = esT , where T is the sample period. Table 8.1 also includes the Laplace transforms, which demonstrates the z = esT relationship for the roots of the denominators of the table entries for F(s) and F(z). Figure 8.4 shows the mapping of lines of constant damping ζ and natural frequency ωn from the s-plane to the upper half of the z-plane, using Eq. (8.10). The mapping has several important features (see Problem 8.4): 1. The stability boundary is the unit circle |z| = 1. 2. The small vicinity around z = +1 in the z-plane is essentially identical to the vicinity around s = 0 in the s-plane. 3. The z-plane locations give response information normalized to the sample rate, rather than to time as in the s-plane. 4. The negative real z-axis always represents a frequency of ωs/2, where ωs = 2π/T = sample rate in radians per second. 5. Vertical lines in the left half of the s-plane (the constant real part or time constant) map into circles within the unit circle of the z-plane. 6. Horizontal lines in the s-plane (the constant imaginary part of the frequency) map into radial lines in the z-plane.

566 Chapter 8 Digital Control Im(z) 1.2

1.0

z  eTs

s  zvn jvn1  z2

T  Sampling period

vn 

3p 5T 0.8

p 2T

2p 5T

7p 10T

3p z0 10T 0.1 0.2

4p 5T

0.6

0.3 1p 5T

0.4 0.5 0.6 0.7

0.4 9p 10T

0.8 0.2

0.9 vn 

0 1.0

p T 0.8

z 1.0 0.6

0.4

0.2

0

0.2

0.4

p 10T 0.6

p 20T 0.8

1.0 Re(z)

Figure 8.4 Natural frequency (solid color) and damping loci (light color) in the z-plane; the portion below the Re(z)-axis (not shown) is the mirror image of the upper half shown Nyquist frequency = ωs /2

7. Frequencies greater than ωs/2, called the Nyquist frequency, appear in the z-plane on top of corresponding lower frequencies because of the circular character of the trigonometric functions imbedded in Eq. (8.10). This overlap is called aliasing or folding. As a result it is necessary to sample at least twice as fast as a signal’s highest frequency component in order to represent that signal with the samples. (We will discuss aliasing in greater detail in Section 8.4.3.) To provide insight into the correspondence between z-plane locations and the resulting time sequence, Fig. 8.5 sketches time responses that would result from poles at the indicated locations. This figure is the discrete companion of Fig. 3.15.

8.2.4 Final Value Theorem The FinalValue Theorem for continuous systems, which we discussed in Section 3.1.6, states that (8.11) lim x(t) = xss = lim sX(s), t→∞

s→0

as long as all the poles of sX(s) are in the left half-plane (LHP). It is often used to find steady-state system errors and/or steady-state gains of portions of a control

8.2 Dynamic Analysis of Discrete Systems

567

Im(z)

1

0.5

0.5

1

Re(z)

Figure 8.5 Time sequences associated with points in the z-plane

Final Value Theorem for discrete systems

system. We can obtain a similar relationship for discrete systems by noting that a constant continuous steady-state response is denoted by X(s) = A/s and leads to the multiplication by s in Eq. (8.11). Therefore, because the constant steady-state response for discrete systems is A , X(z) = 1 − z−1 the discrete Final Value Theorem is lim x(k) = xss = lim (1 − z−1 )X(z)

k→∞

z→1

if all the poles of (1 − z−1 )X(z) are inside the unit circle. For example, to find the DC gain of the transfer function G(z) =

X(z) 0.58(1 + z) = , U(z) z + 0.16

(8.12)

568 Chapter 8 Digital Control we let u(k) = 1 for k ≥ 0, so that U(z) =

1 1 − z−1

and

0.58(1 + z) . (1 − z−1 )(z + 0.16) Applying the Final Value Theorem yields   0.58(1 + z) xss = lim = 1, z→1 z + 0.16 X(z) =

DC gain

so the DC gain of G(z) is unity. To find the DC gain of any stable transfer function, we simply substitute z = 1 and compute the resulting gain. Because the DC gain of a system should not change whether represented continuously or discretely, this calculation is an excellent aid to check that an equivalent discrete controller matches a continuous controller. It is also a good check on the calculations associated with determining the discrete model of a system.

8.3 Design Using Discrete Equivalents Stages in design using discrete equivalents

Design by discrete equivalent, sometimes called emulation, is partially described in Section 4.4 and proceeds through the following stages: 1. Design a continuous compensation as described in Chapters 1 through 7. 2. Digitize the continuous compensation. 3. Use discrete analysis, simulation, or experimentation to verify the design. In Section 4.4 we discussed Tustin’s method for performing the digitization. Armed with an understanding of the z-transform from Section 8.2, we now develop more digitization procedures and analyze the performance of the digitally controlled system. Assume that we are given a continuous compensation D(s) as shown in Fig. 8.1(a). We wish to find a set of difference equations or D(z) for the digital implementation of that compensation in Fig. 8.1(b). First we rephrase the problem as one of finding the best D(z) in the digital implementation shown in Fig. 8.6(a) to match the continuous system represented by D(s) in Fig. 8.6(b). In this section we examine and compare three methods for solving this problem. It is important to remember, as stated earlier, that these methods are approximations; there is no exact solution for all possible inputs because D(s) responds to the complete time history of e(t), whereas D(z) has access to only the samples e(kT ). In a sense, the various digitization techniques simply make different assumptions about what happens to e(t) between the sample points.

Figure 8.6 Comparison of (a) digital and; (b) continuous implementation

e(kT)

e(t)

D(z)

u(kT)

ZOH

u(t)

e(t)

D(s)

T (a)

(b)

u(t)

8.3 Design Using Discrete Equivalents

569

Tustin’s Method As discussed in Section 4.4, one digitization technique is to approach the problem as one of numerical integration. Suppose 1 U(s) = D(s) = , E(s) s which is integration. Therefore,

u(kT ) =

kT −T

e(t) dt +

0

kT kT −T

e(t) dt,

(8.13)

which can be rewritten as u(kT ) = u(kT − T ) + area under e(t) over last T ,

(8.14)

where T is the sample period. For Tustin’s method, the task at each step is to use trapezoidal integration, that is, to approximate e(t) by a straight line between the two samples (Fig. 8.7). Writing u(kT ) as u(k) and u(kT − T ) as u(k − 1) for short, we convert Eq. (8.14) to u(k) = u(k − 1) +

T [e(k − 1) + e(k)], 2

(8.15)

or, taking the z-transform, U(z) T = E(z) 2



1 + z−1 1 − z−1

 =

2 T




1 1−z−1 1+z−1

.

(8.16)

For D(s) = a/(s + a), applying the same integration approximation yields a . D(z) =
−1  2 1−z + a T 1+z−1 In fact, substituting s=

Tustin’s method or bilinear approximation

Figure 8.7 Trapezoidal integration

2 T



1 − z−1 1 + z−1



for every occurrence of s in any D(s) yields a D(z) based on the trapezoidal integration formula. This is called Tustin’s method or the bilinear approximation. Finding Tustin’s approximation by hand for even a simple transfer function requires fairly extensive algebraic manipulations. The c2d function of MATLAB
expedites the process, as shown in the next example. e(t)

kT  T

kT

t

570 Chapter 8 Digital Control

EXAMPLE 8.1

Digital Controller for Example 6.15 Using Tustin’s Approximation Determine the difference equations to implement the compensation from Example 6.15, s/2 + 1 D(s) = 10 , s/10 + 1 at a sample rate of 25 times bandwidth using Tustin’s approximation. Compare the performance against the continuous system and the discrete implementation done in Example 6.15 at a slower sample rate. Solution. The bandwidth (ωBW ) for Example 6.15 is approximately 10 rad/sec, as can be deduced by observing that the crossover frequency (ωc ) is approximately 5 rad/sec and noting the relationship between ωc and ωBW in Fig. 6.51. Therefore, the sample frequency should be ωs = 25 × ωBW = (25)(10) = 250 rad/sec. Normally, when a frequency is indicated with the units of cycles per second, or Hz, it is given the symbol f , so with this convention, we have fs = ωs /(2π)  40 Hz,

(8.17)

and the sample period is then T = 1/fs = 1/40 = 0.025 sec. The discrete compensation is computed by the MATLAB statement sysDs = tf(10*[0.5 1],[0.1 1]); sysDd = c2d(sysDs,0.025,’tustin’); which produces

4.556 − 4.333 z−1 . 1 − 0.7778 z−1 We can then write the difference equation by inspecting Eq. (8.18) to get D(z) = 10

(8.18)

u(k) = 0.7778u(k − 1) + 45.56e(k) − 43.33e(k − 1), or, indexing all time variables by 1, the equivalent is u(k + 1) = 0.7778u(k) + 45.56[e(k + 1) − 0.9510e(k)].

(8.19)

Equation (8.19) computes the new value of the control, u(k + 1), given the past value of the control, u(k), and the new and past values of the error signal, e(k + 1) and e(k). In principle, the difference equation is evaluated initially with k = 0, then k = 1, 2, 3, . . . However, there is usually no requirement that values for all times be saved in memory. Therefore, the computer need only have variables defined for the current and past values. The instructions to the computer to implement the feedback loop in Fig. 8.1(b) with the difference equation from Eq. (8.19) would call for a continual looping through the following code: READ y, r e=r−y  u = 0.7778up + 45.56 e − 0.9510ep

571

Digital controller

1.4 1.2 1 0.8 0.6 0.4 0.2 0

Continuous controller

0

0.2

0.4

0.6

0.8 1 1.2 Time (sec)

1.4

1.6

1.8

2

1.6

1.8

2

(a)

Control, u

Figure 8.8 Comparison between the digital and the continuous controller step response with a sample rate 25 times bandwidth: (a) position; (b) control

Position, y

8.3 Design Using Discrete Equivalents

50 40 30 20 10 0 10 20

Digital controller Continuous controller

0

0.2

0.4

0.6

0.8 1 1.2 Time (sec)

1.4

(b)

OUTPUT u up = u (where up will be the past value for the next loop through) ep = e go back to READ when T sec have elapsed since last READ Use of SIMULINK to compare the two implementations, in a manner similar to that used for Example 6.15, yields the step responses shown in Fig. 8.8. Note that sampling at 25 times bandwidth causes the digital implementation to match the continuous one quite well. Also note that the same case with half the sampling rate whose step response is shown in Fig. 6.59 contains a noticeable degradation in the overshoot (and damping) compared to the continuous case. Generally speaking, if you want to match a continuous system with a digital approximation of the continuous compensation, it is wise to sample at approximately 25 times bandwidth or faster.

8.3.1 Matched Pole–Zero (MPZ) Method Another digitization method, called the matched pole–zero method, is found by extrapolating from the relationship between the s- and z-planes stated in Eq. (8.10). If we take the z-transform of a sampled function x(k), then the poles of X(z) are related to the poles of X(s) according to the relation z = esT . The MPZ technique applies the relation z = esT to the poles and zeros of a transfer function, even though, strictly speaking, this relation applies neither to transfer functions nor even to the zeros of a time sequence. Like all transfer-function digitization methods, the MPZ method is an approximation; here the approximation is motivated partly by the fact that z = esT is the correct s to z transformation for the poles of the transform of a time sequence and

572 Chapter 8 Digital Control partly by the minimal amount of algebra required to determine the digitized transfer function by hand, so as to facilitate checking the computer calculations. Because physical systems often have more poles than zeros, it is useful to arbitrarily add zeros at z = −1, resulting in a 1 + z−1 term in D(z). This causes an averaging of the current and past input values, as in Tustin’s method. We select the low-frequency gain of D(z) so that it equals that of D(s).

MPZ Method Summary 1. Map poles and zeros according to the relation z = esT . 2. If the numerator is of lower order than the denominator, add powers of (z + 1) to the numerator until numerator and denominator are of equal order. 3. Set the DC or low-frequency gain of D(z) equal to that of D(s). The MPZ approximation of D(s) = Kc

s+a s+b

(8.20)

is

z − e−aT , (8.21) z − e−bT where Kd is found by causing the DC gain of D(z) to equal the DC gain of D(s) using the continuous and discrete versions of the Final Value Theorem. The result is a 1 − e−aT Kc = Kd , b 1 − e−bT or   a 1 − e−bT Kd = Kc . (8.22) b 1 − e−aT D(z) = Kd

For a D(s) with a higher-order denominator, Step 2 in the method calls for adding the (z + 1) term. For example, s+a (z + 1)(z − e−aT ) ⇒ D(z) = Kd , s(s + b) (z − 1)(z − e−bT ) where, after dropping the poles at s = 0 and z = 1,   a 1 − e−bT Kd = Kc . 2b 1 − e−aT D(s) = Kc

(8.23)

(8.24)

In the digitization methods described so far, the same power of z appears in the numerator and denominator of D(z). This implies that the difference equation output at time k will require a sample of the input at time k. For example, the D(z) in Eq. (8.21) can be written U(z) 1 − αz−1 = D(z) = Kd , E(z) 1 − βz−1

(8.25)

where α = e−aT and β = e−bT . By inspection we can see that Eq. (8.25) results in the difference equation u(k) = βu(k − 1) + Kd [e(k) − αe(k − 1)].

(8.26)

8.3 Design Using Discrete Equivalents

EXAMPLE 8.2

573

Design of a Space Station Attitude Digital Controller Using Discrete Equivalents A very simplified model of the space station attitude control dynamics has the plant transfer function 1 G(s) = 2 . s Design a digital controller to have a closed-loop natural frequency ωn ∼ = 0.3 rad/sec and a damping ratio ζ = 0.7.

Figure 8.9 Continuous-design definition for Example 8.2



R




E

D(s)

U



1 s2

Y

Solution. The first step is to find the proper D(s) for the system defined in Fig. 8.9. After some trial and error, we find that the specifications can be met by the lead compensation s + 0.2 D(s) = 0.81 . (8.27) s+2 The root locus in Fig. 8.10 verifies the appropriateness of using Eq. (8.27). To digitize this D(s), we first need to select a sample rate. For a system with ωn = 0.3 rad/sec, the bandwidth will also be about 0.3 rad/sec, and an acceptable sample rate would be about 20 times ωn . Thus ωs = 0.3 × 20 = 6 rad/sec. A sample rate of 6 rad/sec is about 1 Hertz; therefore, the sample period should be T = 1 sec. The MPZ digitization of Eq. (8.27), given by Eqs. (8.21) and (8.22), yields z − 0.82 D(z) = 0.389 z − 0.135 0.389 − 0.319z−1 . 1 − 0.135z−1 Inspection of Eq. (8.28) gives us the difference equation =

u(k) = 0.135u(k − 1) + 0.389e(k) − 0.319e(k − 1), Figure 8.10 s-Plane locus with respect to K

Im(s) 0.2

2 Selected roots (Kc  0.81)

0.2

Re(s)

(8.28)

(8.29)

574 Chapter 8 Digital Control Figure 8.11 A digital control system that is equivalent to Fig. 8.9

Computer r(k)

u(k)

Eq. (8.29)

ZOH

u(t)

Plant 1 s2

y(t)

Sampler (T  1 sec) and analog-to-digital conversion

y(k)

where e(k) = r(k) − y(k), and this completes the digital algorithm design. The complete digital system is shown in Fig. 8.11. The last step in the design process is to verify the design by implementing it on the computer. Figure 8.12 compares the step response of the digital system using T = 1 sec with the step response of the continuous compensation. Note that there is greater overshoot and a longer settling time in the digital system, which suggests a decrease in the damping. The average T/2 delay shown in Fig. 8.2 is the cause of the reduced damping. For a better match to the continuous system, it may be prudent to increase the sample rate. Figure 8.12 also shows the response with sampling that is twice as fast and it can be seen that it comes much closer to the continuous system. Note that the discrete compensation needs to be recalculated for this faster sample rate according to Eqs. (8.21) and (8.22). It is impossible to sample e(k), compute u(k), and then output u(k) all in zero elapsed time; therefore, Eqs. (8.26) and (8.29) are impossible to implement precisely. However, if the equation is simple enough and/or the computer is fast enough, a slight computational delay between the e(k) sample and the u(k) output will have a Figure 8.12 Step responses of the continuous and digital implementations

1.6 1.4 Plant output

1.2 1.0 0.8 0.6 0.4 0.2 0

0

5

10

15 Time (sec)

20

25

Continuous design Discrete equivalent design, T  1 sec Discrete equivalent design, T  0.5 sec

30

8.3 Design Using Discrete Equivalents

575

negligible effect on the actual response of the system compared with that expected from the original design. A rule of thumb would be to keep the computational delay on the order of 1/10 of T . The real-time code and hardware can be structured so that this delay is minimized by making sure that computations between read A/D and write D/A are minimized and that u(k) is sent to the ZOH immediately after its calculation.

8.3.2 Modified Matched Pole–Zero (MMPZ) Method The D(z) in Eq. (8.23) would also result in u(k) being dependent on e(k), the input at the same time point. If the structure of the computer hardware prohibits this relation or if the computations are particularly lengthy, it may be desirable to derive a D(z) that has one less power of z in the numerator than in the denominator; hence, the computer output u(k) would require only input from the previous time, that is, e(k − 1). To do this, we simply modify Step 2 in the matched pole–zero procedure so that the numerator is of lower order than the denominator by 1. For example, if D(s) = Kc we skip Step 2 to get

s+a , s(s + b)

z − e−aT , (z − 1)(z − e−bT )   a 1 − e−bT . Kd = Kc b 1 − e−aT

D(z) = Kd

(8.30)

To find the difference equation, we multiply the top and bottom of Eq. (8.30) by z−2 to obtain z−1 (1 − e−aT z−1 ) D(z) = Kd . (8.31) 1 − z−1 (1 + e−bT ) + z−2 e−bT By inspecting Eq. (8.31) we can see that the difference equation is u(k) = (1 + e−bT )u(k − 1) − e−bT u(k − 2) + Kd [e(k − 1) − e−aT e(k − 2)]. In this equation an entire sample period is available to perform the calculation and to output u(k), because it depends only on e(k − 1). A discrete analysis of this controller would therefore more accurately explain the behavior of the actual system. However, because this controller is using data that are one cycle old, it will typically not perform as well as the MPZ controller in terms of the deviations of the desired system output in the presence of random disturbances.

8.3.3 Comparison of Digital Approximation Methods A numerical comparison of the magnitude of the frequency response for a firstorder lag, 5 , D(s) = s+5 is made in Fig. 8.13 for the three approximation techniques at two different sample rates. The results of the D(z) computations used in Fig. 8.13 are shown in Table 8.2.

576 Chapter 8 Digital Control T  1 15 sec vs  100 rad/sec

T  1 3 sec vs  20 rad/sec

1

1

MMPZ Tustin's,

D(z)

MMPZ Tustin’s, MPZ

0.1

MPZ

D(z) 0.1

D(s) 0.5

5

D(s)

50 v (rad/sec)

0.5

5

50

v (rad/sec)

Figure 8.13 A comparison of the frequency response of three discrete approximations

Figure 8.13 shows that all the approximations are quite good at frequencies below about 1/4 the sample rate, or ωs/4. If ωs/4 is sufficiently larger than the filter break-point frequency—that is, if the sampling is fast enough—the break-point characteristics of the lag will be accurately reproduced. Tustin’s technique and the MPZ method show a notch at ωs/2 because of their zero at z = −1 from the z + 1 term. Other than the large difference at ωs/2, which is typically outside the range of interest, the three methods have similar accuracies.

8.3.4 Applicability Limits of the Discrete Equivalent Design Method

ZOH transfer function

If we performed an exact discrete analysis or a simulation of a system and determined the digitization for a wide range of sample rates, the system would often be unstable for rates slower than approximately 5ωn , and the damping would be degraded significantly for rates slower than about 10ωn . At sample rates  20ωn (or  20 times the bandwidth for more complex systems), design by discrete equivalents yields reasonable results, and at sample rates of 30 times the bandwidth or higher, discrete equivalents can be used with confidence. As shown by Fig. 8.2, the errors come about because the technique ignores the lagging effect of the ZOH which, on the average, is T/2. A method to account for

TABLE 8.2

Comparing Digital Approximations of D(z) for D(s) = 5/(s + 5) ωs Method Matched pole–zero (MPZ) Modified MPZ (MMPZ) Tustin’s

100 rad/sec z+1 z − 0.715 1 0.285 z − 0.715 z+1 0.143 z − 0.713 0.143

20 rad/sec z+1 z − 0.189 1 0.811 z − 0.189 z+1 0.454 z − 0.0914 0.405

8.4 Hardware Characteristics

577

this is to approximate the T/2 delay with Eq. (5.94) by including a transfer function approximation for the ZOH:1 2/T GZOH (s) = . (8.32) s + 2/T Once an initial design is carried out and the sampling rate has been selected, we could improve on our discrete design by inserting Eq. (8.32) into the original plant model and adjusting the D(s) so that a satisfactory response in the presence of the sampling delay is achieved. Therefore, we see that use of Eq. (8.32) partially alleviates the approximate nature of using discrete equivalents. For sample rates slower than about 10ωn it is advisable to analyze the entire system using an exact discrete analysis. If a discrete analysis shows an unacceptable degradation of performance due to the sampling, the design can then be refined using exact discrete methods. We cover this approach in Section 8.6.

8.4 Hardware Characteristics A digital control system includes several unique components not found in continuous control systems: an analog-to-digital converter is a device to sample the continuous signal voltage from the sensor and to convert that signal to a digital word; a digitalto-analog converter is a device to convert the digital word from the computer to an analog voltage, an anti-alias prefilter is an analog device designed to reduce the effects of aliasing, and the computer is the device where the compensation D(z) is programmed and the calculations are carried out. This section provides a brief description of each of these.

8.4.1 Analog-to-Digital (A/D) Converters As discussed in Section 8.1, A/D converters are devices that convert a voltage level from a sensor to a digital word usable by the computer. At the most basic level, all digital words are binary numbers consisting of many bits that are set to either 1 or 0. Therefore, the task of the A/D converter at each sample time is to convert a voltage level to the correct bit pattern and often to hold that pattern until the next sample time. Of the many A/D conversion techniques that exist, the most common are based on counting schemes or a successive-approximation technique. In counting methods the input voltage may be converted to a train of pulses whose frequency is proportional to the voltage level. The pulses are then counted over a fixed period using a binary counter, thus resulting in a binary representation of the voltage level. A variation on this scheme is to start the count simultaneously with a voltage that is linear in time and to stop the count when the voltage reaches the magnitude of the input voltage to be converted. The successive-approximation technique tends to be much faster than the counting methods. It is based on successively comparing the input voltage to reference levels representing the various bits in the digital word. The input voltage is first compared with a reference value that is half the maximum. If the input voltage is greater, the most significant bit is set, and the signal is then compared with a reference level 1 Or other Padé approximate as discussed in Section 5.6.3

578 Chapter 8 Digital Control that is 3/4 the maximum to determine the next bit, and so on. One clock cycle is required to set each bit, so an n-bit converter would require n cycles. At the same clock rate a counter-based converter might require as many as 2n cycles, which would usually be much slower. With either technique, the greater the number of bits, the longer it will take to perform the conversion. The price of A/D converters generally goes up with both speed and bit size. In 2009, a 14-bit (resolution of 0.006%) converter with a high performance capability of a 10-n sec conversion time (100 million samples per sec) sold for approximately $25 while a 12-bit (0.025%) converter with a good performance capability of a 1 µ sec conversion time (1 million samples per sec) sold for approximately $4. An 8-bit (0.4% resolution) with a 1 µ sec conversion time sold for approximately $1. The performance has been improving considerably every year. If more than one channel of data needs to be sampled and converted to digital words, it is usually accomplished by use of a multiplexer rather than by multiple A/D converters. The multiplexer sequentially connects the converter into the channel being sampled.

8.4.2 Digital-to-Analog (D/A) Converters D/A converters, as mentioned in Section 8.1, are used to convert the digital words from the computer to a voltage level and are sometimes referred to as Sample and Hold devices. They provide analog outputs from a computer for driving actuators or perhaps a recording device such as an oscilloscope or strip-chart recorder. The basic idea behind their operation is that the binary bits cause switches (electronic gates) to open or close, thus routing the electric current through an appropriate network of resistors to generate the correct voltage level. Because no counting or iteration is required for such converters, they tend to be much faster than A/D converters. In fact, A/D converters that use the successive-approximation method of conversion include D/A converters as components.

8.4.3 Anti-Alias Prefilters Analog prefilters reduce aliasing

Nyquist–Shannon sampling theorem

An analog anti-alias prefilter is often placed between the sensor and the A/D converter. Its function is to reduce the higher-frequency noise components in the analog signal in order to prevent aliasing, that is, having the noise be modulated to a lower frequency by the sampling process. An example of aliasing is shown in Fig. 8.14, where a 60 Hertz oscillatory signal is being sampled at 50 Hertz. The figure shows the result from the samples as a 10 Hertz signal and also shows the mechanism by which the frequency of the signal is aliased from 60 to 10 Hertz. Aliasing will occur any time the sample rate is not at least twice as fast as any of the frequencies in the signal being sampled. Therefore, to prevent aliasing of a 60 Hertz signal, the sample rate would have to be faster than 120 Hertz, clearly much higher than the 50 Hertz rate in the figure. Aliasing is one of the consequences of the sampling theorem of Nyquist and Shannon. Their theorem basically states that, for the signal to be accurately reconstructed from the samples, it must have no frequency component greater than half the sample rate ( ωs/2). Another consequence of their theorem is that the highest frequency

8.4 Hardware Characteristics Figure 8.14 An example of aliasing

Aliased signal from samples

579

60-Hertz signal

0.02

0.08

0.1

Time (sec)

50-Hertz samples

that can be unambiguously represented by discrete samples is the Nyquist rate of ωs/2, an idea we discussed in Section 8.2.3. The consequence of aliasing on a digital control system can be substantial. In a continuous system, noise components with a frequency much higher than the controlsystem bandwidth normally have a small effect because the system will not respond at the high frequency. However, in a digital system, the frequency of the noise can be aliased down to the vicinity of the system bandwidth so that the closed-loop system would respond to the noise. Thus, the noise in a poorly designed digitally controlled system could have a substantially greater effect than if the control had been implemented using analog electronics. The solution is to place an analog prefilter before the sampler. In many cases a simple first-order low-pass filter will do—that is, Hp (s) =

a , s+a

where the break point a is selected to be lower than ωs/2 so that any noise present with frequencies greater than ωs/2 is attenuated by the prefilter. The lower the breakpoint frequency selected, the more the noise above ωs/2 is attenuated. However, too low a break point may force the designer to reduce the control system’s bandwidth. The prefilter does not completely eliminate the aliasing; however, through judicious choice of the prefilter break point and the sample rate, the designer has the ability to reduce the magnitude of the aliased noise to some acceptable level.

8.4.4 The Computer The computer is the unit that does all the computations. Most digital controllers used today are built around a microcontroller that contains both a microprocessor and most of the other functions needed, including theA/D and D/A conversion. For development purposes in a laboratory, a digital controller could be a desktop-sized workstation or a PC. The relatively low cost of microprocessor technology has accounted for the large increase in the use of digital control systems, which started in the 1980s and continues into the 2000s. The computer consists of a central processor unit (CPU), which does the computations and provides the system logic; a clock to synchronize the system; memory

580 Chapter 8 Digital Control modules for data and instruction storage; and a power supply to provide the various required voltages. The memory modules come in three basic varieties: 1. Read-only memory (ROM) is the least expensive, but after its manufacture its contents cannot be changed. Most of the memory in products manufactured in quantity is ROM. It retains its stored values when power is removed. 2. Random-access memory (RAM) is the most expensive, but its values can be changed by the CPU. It is required only to store the values that will be changed during the control process and typically represents only a small fraction of the total memory of a developed product. It loses the values in memory when power is removed. 3. Programmable read-only memory (EPROM) is a ROM whose values can be changed by a technician using a special device. It is typically used during product development to enable the designer to try different algorithms and parameter values. It retains its stored values when power is removed. In some products, it is useful to have a few of the stored quantities in EPROMs so that individual calibrations can be carried out for each unit. Microprocessors for control applications generally come with a digital word size of 8, 16, or 32 bits, although some have been available with 12 bits. Larger word sizes give better accuracy, but at an increase in cost. The most economical solution is often to use an 8-bit microprocessor, but to use two digital words to store one value (double precision) in the areas of the controller that are critical to the system accuracy. Many digital control systems use computers originally designed for digital signal-processing applications, so-called DSP chips.

8.5 Sample-Rate Selection The selection of the best sample rate for a digital control system is the result of a compromise of many factors. Sampling too fast can cause a loss of accuracy while the basic motivation for lowering the sample rate ωs is cost. A decrease in sample rate means more time is available for the control calculations; hence slower computers can be used for a given control function or more control capability can be achieved from a given computer. Either way, the cost per function is lowered. For systems with A/D converters, less demand on conversion speed will also lower cost. These economic arguments indicate that the best engineering choice is the slowest possible sample rate that still meets all performance specifications. There are several factors that could provide a lower limit on the acceptable sample rate: 1. tracking effectiveness as measured by closed-loop bandwidth or by time-response requirements, such as rise time and settling time; 2. regulation effectiveness as measured by the error response to random plant disturbances; 3. error due to measurement noise and the associated prefilter design methods. A fictitious limit occurs when using discrete equivalents. The inherent approximation in the method may give rise to decreased performance or even system instabilities

8.5 Sample-Rate Selection

581

as the sample rate is lowered. This can lead the designer to conclude that a faster sample rate is required. However, there are two solutions: 1. sample faster, and 2. recognize that the approximations are invalid and refine the design with a direct digital-design method described in the subsequent sections. The ease of designing digital control systems with fast sample rates and the low cost of very capable computers often drives the designer to select a sample rate that is 40 × ωBW or higher. For computers with fixed-point arithmetic, very fast sample rates can lead to multiplication errors that have the potential to produce significant offsets or limit cycles in the control (see Franklin et al., 1998).

8.5.1 Tracking Effectiveness An absolute lower bound on the sample rate is set by a specification to track a command input with a certain frequency (the system bandwidth). The sampling theorem (see Section 8.4.3 and Franklin et al., 1998) states that in order to reconstruct an unknown, band-limited, continuous signal from samples of that signal, we must sample at least twice as fast as the highest frequency contained in the signal. Therefore, in order for a closed-loop system to track an input at a certain frequency, it must have a sample rate twice as fast; that is, ωs must be at least twice the system bandwidth (ωs ⇒ 2 × ωBW ). We also saw from the results of mapping the s-plane into the z-plane (z = esT ) that the highest frequency that can be represented by a discrete system is ωs/2, which supports the conclusion of the theorem. It is important to note the distinction between the closed-loop bandwidth ωBW and the highest frequency in the open-loop plant dynamics, because the two frequencies can be quite different. For example, closed-loop bandwidths can be an order of magnitude less than open-loop modes of resonances for some control problems. Information concerning the state of the plant resonances for purposes of control can be extracted from sampling the output without satisfying the sampling theorem because some a priori knowledge concerning these dynamics (albeit imprecise) is available, and the system is not required to track these frequencies. Thus a priori knowledge of the dynamic model of the plant can be included in the compensation in the form of a notch filter. The closed-loop-bandwidth limitation provides the fundamental lower bound on the sample rate. In practice, however, the theoretical lower bound of sampling at twice the bandwidth of the reference input signal would not be judged sufficient in terms of the quality of the desired time responses. For a system with a rise time on the order of 1 sec (thus yielding a closed-loop bandwidth on the order of 0.5 Hertz), it is reasonable to insist on a sampling rate of 10 to 20 Hertz, which is a factor of 20 to 40 times ωBW . The purposes of choosing a sample rate much greater than the bandwidth are to reduce the delay between a command and the system response to the command and also to smooth the system output to the control steps coming out of the ZOH.

8.5.2 Disturbance Rejection Disturbance rejection is an important—if not the most important—aspect of any control system. Disturbances enter a system with various frequency characteristics

582 Chapter 8 Digital Control ranging from steps to white noise. For the purpose of sample-rate selection, the higher-frequency random disturbances are the most influential. The ability of the control system to reject disturbances with a good continuous controller represents the lower bound on the error response that we can hope for when implementing the controller digitally. In fact, some degradation relative to the continuous design must occur because the sampled values are slightly out of date at all times except precisely at the sampling instants. However, if the sample rate is very fast compared with the frequencies contained in the noisy disturbance, we should expect no appreciable loss from the digital system as compared with the continuous controller. At the other extreme, if the sample rate is very slow compared with the characteristic frequencies of the noise, the response of the system because of noise is essentially the same as the response we would get if the system had no control at all. The selection of a sample rate will place the response somewhere in between these two extremes. Thus, the impact of the sample rate on the ability of the system to reject disturbances may be very important to consider when choosing the sample rate. Although the best choice of sample rate in terms of the ωBW multiple is dependent on the frequency characteristics of the noise and the degree to which random disturbance rejection is important to the quality of the controller, sample rates on the order of 25 times ωBW or higher are typical.

8.5.3 Effect of Anti-Alias Prefilter Digital control systems with analog sensors typically include an analog anti-alias prefilter between the sensor and the sampler as described in Section 8.4.3. The prefilters are low-pass, and the simplest transfer function is Hp (s) =

a , s+a

so that the noise above the prefilter break point a is attenuated. The goal is to provide enough attenuation at half the sample rate ( ωs/2) that the noise above ωs/2, when aliased into lower frequencies by the sampler, will not be detrimental to control system performance. A conservative design procedure is to select ωs and the break point to be sufficiently higher than the system bandwidth that the phase lag from the prefilter does not significantly alter the system stability. This would allow the prefilter to be ignored in the basic control system design. Furthermore, for a good reduction in the highfrequency noise at ωs/2, we choose a sample rate that is about 5 or 10 times higher than the prefilter break point. The implication of this prefilter design procedure is that sample rates need to be on the order of 30 to 100 times faster than the system bandwidth. Using this conservative design procedure, the prefilter influence will likely provide the lower bound on the selection of the sample rate. An alternative design procedure is to allow significant phase lag from the prefilter at the system bandwidth. This requires us to include the analog prefilter characteristics in the plant model when carrying out the control design. It allows the use of lower sample rates, but at the possible expense of increased complexity in the compensation because additional phase lead must be provided to counteract the prefilter’s phase lag. If this procedure is used and low prefilter break points are allowed, the effect of

8.6 Discrete Design

583

sample rate on sensor noise is small, and the prefilter essentially has no effect on the sample rate. It may seem counterintuitive that placing a lag (the analog prefilter) in one portion of the controller and a counteracting lead [extra lead in D(z)] in another portion of the controller provides a net positive effect on the overall system. The net gain is a result of the fact that the lag is in the analog part of the system where high frequencies can exist. The counteracting lead is in the digital part of the system where frequencies above the Nyquist rate do not exist. The result is a reduction in the high frequencies before the sampling which are not reamplified by the counteracting digital lead, thus producing net reduction in high frequencies. Furthermore, these high frequencies are particularly insidious with a digital controller because of the aliasing that would result from the sampling.

8.5.4 Asynchronous Sampling As noted in the previous paragraphs, divorcing the prefilter design from the controllaw design may require using a faster sample rate than otherwise. This same result may show up in other types of architecture. For example, a smart sensor with its own computer running asynchronously relative to the primary control computer will not be amenable to direct digital design because the overall system transfer function depends on the phasing between the smart sensor and the primary digital controller. This situation is similar to that of the digitization errors discussed in Section 8.6. Therefore, if asynchronous digital subsystems are present, sample rates on the order of 20 × ωBW or slower in any module should be used with caution and the system performance checked through simulation or experiment.


8.6 Discrete Design It is possible to obtain an exact discrete model that relates the samples of the continuous plant y(k) to the input control sequence u(k). This plant model can be used as part of a discrete model of the feedback system including the compensation D(z). Analysis and design using this discrete model is called discrete design or, alternatively, direct digital design. The following subsections will describe how to find the discrete plant model (Section 8.6.1), what the feedback compensation looks like when designing with a discrete model (Sections 8.6.2 and 8.6.3), and how the design process is carried out (Section 8.6.4).

8.6.1 Analysis Tools

The exact discrete equivalent

The first step in performing a discrete analysis of a system with some discrete elements is to find the discrete transfer function of the continuous portion. For a system similar to that shown in Fig. 8.1(b), we wish to find the transfer function between u(kT ) and y(kT ). Unlike the cases discussed in the previous sections, there is an exact discrete equivalent for this system, because the ZOH precisely describes what happens between samples of u(kT ) and the output y(kT ) is dependent only on the input at the sample times u(kT ).

584 Chapter 8 Digital Control R(s)






ZOH

D(z)

G(s)



Y(s) R(z)






D(z)

G(z)

Y(z)



(b)

(a)

Figure 8.15 Comparison of (a) a mixed control system; and (b) its pure discrete equivalent

For a plant described by G(s) and preceded by a ZOH, the discrete transfer function is   G(s) −1 G(z) = (1 − z )Z , (8.33) s where Z{F(s)} is the z-transform of the sampled time series whose Laplace transform is the expression for F(s), given on the same line in Table 8.1. Equation (8.33) has the term G(s)/s because the control comes in as a step input from the ZOH during each sample period. The term 1 − z−1 reflects the fact that a one-sample duration step can be thought of as an infinite duration step followed by a negative step one cycle delayed. For a more complete derivation, see Franklin et al. (1998). Equation (8.33) allows us to replace the mixed (continuous and discrete) system shown in Fig. 8.15(a) with the equivalent pure discrete system shown in Fig. 8.15(b). The analysis and design of discrete systems is very similar to the analysis and design of continuous systems; in fact, all the same rules apply. The closed-loop transfer function of Fig. 8.15(b) is obtained using the same rules of block-diagram reduction—that is, Y (z) D(z)G(z) = . (8.34) R(z) 1 + D(z)G(z) To find the characteristic behavior of the closed-loop system, we need to find the factors in the denominator of Eq. (8.34)—that is, the roots of the discrete characteristic equation 1 + D(z)G(z) = 0. The root-locus techniques used in continuous systems to find roots of a polynomial in s apply equally well and without modification to the polynomial in z; however, the interpretation of the results is quite different, as we saw in Fig. 8.4. A major difference is that the stability boundary is now the unit circle instead of the imaginary axis.

EXAMPLE 8.3

Discrete Root Locus For the case in which G(s) in Fig. 8.15(a) is G(s) =

a s+a

and D(z) = K, draw the root locus with respect to K, and compare your results with a root locus of a continuous version of the system. Discuss the implications of your loci.

8.6 Discrete Design Figure 8.16 Root loci for (a) the z-plane; and (b) the s-plane

Im(z)

z  1

585

Im(s)

za

z0

Re(z)

s  a

(a)

Re(s)

(b)

Solution. It follows from Eq. (8.33) that   a G(z) = (1 − z−1 )Z s(s + a)   (1 − e−aT )z−1 −1 = (1 − z ) (1 − z−1 )(1 − e−aT z−1 ) =

1−α , z−α

where

α = e−aT .

To analyze the performance of the closed-loop system, standard root-locus rules apply. The result is shown in Fig. 8.16(a) for the discrete case and in Fig. 8.16(b) for the continuous case. In contrast to the continuous case, in which the system remains stable for all values of K, in the discrete case the system becomes oscillatory with decreasing damping ratio as z goes from 0 to −1 and eventually becomes unstable. This instability is due to the lagging effect of the ZOH, which is properly accounted for in the discrete analysis.

8.6.2 Feedback Properties

Discrete control laws

In continuous systems we typically start the design process by using the following basic design elements: proportional, derivative, or integral control laws, or some combination of these, sometimes with a lag included. The same ideas can be used in discrete design. Alternatively, the D(z) resulting from the digitization of a continuously designed D(s) will produce these basic design elements, which will then be used as a starting point in a discrete design. The discrete control laws are as follows:

Proportional u(k) = Ke(k) ⇒ D(z) = K.

(8.35)

u(k) = KTD [e(k) − e(k − 1)],

(8.36)

Derivative

586 Chapter 8 Digital Control for which the transfer function is D(z) = KTD (1 − z−1 ) = KTD

z−1 z−1 = kD . z z

(8.37)

Kp e(k), TI

(8.38)

Integral u(k) = u(k − 1) +

for which the transfer function is       K 1 K z z D(z) = = = k . I TI 1 − z−1 TI z − 1 z−1

(8.39)

Lead Compensation The examples in Section 8.3 showed that a continuous lead compensation leads to difference equations of the form u(k + 1) = βu(k) + K[e(k + 1) − αe(k)],

(8.40)

for which the transfer function is D(z) = K

1 − αz−1 . 1 − βz−1

(8.41)

8.6.3 Discrete Design Example Digital control design consists of using the basic feedback elements of Eqs. (8.35) to (8.41) and iterating on the design parameters until all specifications are met.

EXAMPLE 8.4

Direct Discrete Design of the Space Station Digital Controller Design a digital controller to meet the same specifications as in Example 8.2 using discrete design. Solution. The discrete model of the 1/s2 plant, preceded by a ZOH, is found through Eq. (8.33) to be   z+1 T2 G(z) = , 2 (z − 1)2 which, with T = 1 sec, becomes   z+1 1 . G(z) = 2 (z − 1)2 Proportional feedback in the continuous case yields pure oscillatory motion, so in the discrete case we should expect even worse results. The root locus in Fig. 8.17 verifies this. For very low values of K (where the locus represents roots at very low frequencies compared to the sample rate), the locus is tangent to the unit circle (ζ ∼ = 0 indicating pure oscillatory motion), thus matching the proportional continuous design. For higher values of K, Fig. 8.17 shows that the locus diverges into the unstable region because of the effect of the ZOH and sampling. To compensate for this, we will add a derivative term to the proportional term so that the control law is U(z) = K[1 + TD (1 − z−1 )]E(z),

(8.42)

8.6 Discrete Design Figure 8.17 z-plane root locus for a 1/s2 plant with proportional feedback

587

Im(z)

1

Re(z)

which yields compensation of the form

z−α , (8.43) z where the new K and α replace the K and TD in Eq. (8.42). Now the task is to find the values of α and K that yield good performance. The specifications for the design are that ωn = 0.3 rad/sec and ζ = 0.7. Figure 8.4 indicates that this s-plane root location maps into a desired z-plane location of z = 0.78 ± 0.18j. Figure 8.18 is the locus with respect to K for α = 0.85. The location of the zero (at z = 0.85) was determined by trial and error until the locus passed through the desired z-plane location. The value of the gain when the locus passes through z = 0.78±0.18j is K = 0.374. Equation (8.43) now becomes z − 0.85 . (8.44) D(z) = 0.374 z Normally, it is not particularly advantageous to match specific z-plane root locations; rather it is necessary only to pick K and α (or TD ) to obtain acceptable z-plane roots, a much easier task. In this example, we want to match a specific location only so that we can compare the result with the design in Example 8.2. The control law that results is U(z) = 0.374(1 − 0.85z−1 )E(z), or u(k) = 0.374e(k) − 0.318e(k − 1), (8.45) which is similar to the control equation (8.29) obtained previously. D(z) = K

The controller in Eq. (8.45) basically differs from the continuously designed controller [Eq. (8.29)] only in the absence of the u(k − 1) term. The u(k − 1) term in Eq. (8.29) results from the lag term (s + b) in the compensation [Eq. (8.27)]. The lag term is typically included in analog controllers both because it supplies noise attenuation and because pure analog differentiators are difficult to build. Some equivalent lag in discrete design naturally appears as a pole at z = 0 (see Fig. 8.18) and represents the one-sample delay in computing the derivative by a first difference. For more noise attenuation, we could move the pole to the right of z = 0, thus resulting in less derivative action and more smoothing, the same trade-off that exists in continuous control design.

588 Chapter 8 Digital Control Figure 8.18 z-plane locus for the 1/s2 plant with D(z) = K(z − 0.85)/z

1.0 Desired root

0.8 0.6

Imaginary axis

0.4 0.2

0.2 0.4 0.6 0.8 1.0 1.0

Desired root 0.5

0

0.5

1.0

Real axis

8.6.4 Discrete Analysis of Designs Any digital controller, whether designed by discrete equivalents or directly in the z-plane, can be analyzed using discrete analysis, which consists of the following steps: 1. Find the discrete model of the plant and ZOH using Eq. (8.33). 2. Form the feedback system including D(z). 3. Analyze the resulting discrete system. We can determine the roots of the system using a root locus, as described in Section 8.6.3, or we can determine the time history (at the sample instants) of the discrete system.

EXAMPLE 8.5

Damping and Step Response in Digital versus Continuous Design Use discrete analysis to determine the equivalent s-plane damping and the step responses of the digital designs in Examples 8.2 and 8.4, and compare your results with the damping and step response of the continuous case in Example 8.2. Solution. The MATLAB statements to evaluate the damping and step response of the continuous case in Example 8.2 are sysGs = tf(1,[1 0 0]); sysDs = tf(0.81 * [1 0.2],[1 2]); sysGDs = series(sysGs,sysDs); sysCLs = feedback(sysGDs,1,1); step(sysCLs) damp(sysCLs)

8.6 Discrete Design

589

To analyze the digital control cases, the model of the plant preceded by the ZOH is found using the statements T = 1; sysGz = c2d(sysGs,T,‘zoh’) Analysis of the digital control designed using the discrete equivalent [Eq. (8.29)] in Example 8.2 is performed by the statements sysDz = tf( [.389 −.319],[1 −.135]) sysDGz = series(sysGz,sysDz) sysCLz = feedback(sysDGz,1) step(sysCLz,T) damp(sysCLz,T) Likewise, the discrete design of D(z) from Eq. (8.44) can be analyzed by the same sequence. The resulting step responses are shown in Fig. 8.19. The calculated damping ζ and complex root natural frequencies ωn of the closed-loop systems are ζ = 0.705,

ωn = 0.324;

Discrete equivalent: ζ = 0.645,

ωn = 0.441;

ζ = 0.733,

ωn = 0.306.

Continuous case: Discrete design:

The figure shows increased overshoot for the discrete equivalent method that occurred because of the decreased damping of that case. Very little increased overshoot occurred in the discrete design, because that compensation was adjusted specifically so that the equivalent s-plane damping of the discrete system was approximately at the desired damping value of ζ = 0.7.

1.6 1.4 1.2 Plant output

Figure 8.19 Step response of the continuous and digital systems in Examples 8.2 and 8.4

1.0 0.8 0.6 0.4 0.2 0

0

5

10

15 Time (sec)

20

Continuous design Discrete equivalent design Discrete design

25

30

590 Chapter 8 Digital Control Although the analysis showed some differences between the performance of the digital controllers designed by the two methods, neither the performance nor the control equations [Eqs. (8.29) and (8.45)] are very different. This similarity results because the sample rate is fairly fast compared with ωn —that is, ωs ∼ = 20 × ωn . If we were to decrease the sample rate, the numerical values in the compensations would become increasingly different and the performance would degrade considerably for the discrete equivalent case. As a general rule, discrete design should be used if the sampling frequency is slower than 10 × ωn . At the very least, a discrete equivalent design with slow sampling (ωs < 10 × ωn ) should be verified by a discrete analysis or by simulation, as described in Section 4.4, and the compensation adjusted if needed. A simulation of a digital control system is a good idea in any case. If it properly accounts for all delays and possibly asynchronous behavior of different modules, it may expose instabilities that are impossible to detect using continuous or discrete linear analysis. A more complete discussion regarding the effects of sample rate on the design is contained in Section 8.5.

8.7 Historical Perspective One of the earliest examples of actual control of systems based on sampled data came with the use of search RADAR in WWII. In that case, the position of the target was available only once each revolution of the antenna. The theory of sampled data systems was developed by the mathematician W. Hurewicz2 and published as a chapter in H. M. James, N. B. Nichols, and R. S. Phillips, Theory of Servomechanisms, vol. 25, Rad Lab Series, New York, McGraw Hill, 1947. The historical perspective for Chapter 5 discussed the introduction of computers for engineers performing design activities. The possibility of using computers for direct digital control motivated the continuation of work on sampled data systems during the 1950s, especially at Columbia University under Professor J. R. Ragazzini. That work was published in J. R. Ragazzini and G. F. Franklin, Sampled-Data Control Systems, New York, McGraw Hill, 1958. Early applications were in the process control industry where the relatively large and expensive computers available at the time could be justified. Professor Karl Astrom introduced direct digital control of a paper mill in Sweden in the early 1960s. In 1961, when President Kennedy announced the goal of sending a man to the moon, there were no digital autopilots for aerospace vehicles. In fact, small digital computers suitable for implementing control systems were virtually nonexistent. The team at the MIT Draper Labs (called the Instrumentation Lab at that time) in charge of designing and building the Apollo control systems initially designed the control systems for the lunar and command modules with conventional analog electronics. However, they discovered that those systems would be too heavy and complex for the mission. So the decision was made to design and build the first aerospace digital 2 Hurewicz died in 1956 falling off a ziggurat (a Mexican pyramid) on a conference outing at the Inter-

national Symposium on algebraic topology in Mexico. It is suggested that he was: “... a paragon of absentmindedness, a failing that probably led to his death.”

Summary

591

control system. Bill Widnall, Dick Battin, and Don Fraser were all key players in the successful design and execution of that system for the Apollo flights in late 1960s. The group went on to demonstrate a digital autopilot for NASA’s F-8 in the 1970s, and digital autopilots went on to become dominant over the 1980s and beyond. In fact, with the introduction of inexpensive digital signal processors, most control systems of any kind became digital by the turn of the century and, today, very few control systems are being implemented with analog electronics. This evolution has had an effect on the training for controls engineers. In the past, the ability to design and build the specialized circuitry for analog electronic controls caused many controls engineers to have an Electrical Engineering background. Now, with easily programmable digital computers being readily available, the background of controls engineers tends more toward the specialties that are most familiar with the systems being controlled.

SUMMARY • •

•

The simplest and most expedient design technique is to transform a continuous controller design to its discrete form—that is, to use its discrete equivalent. Design using discrete equivalents entails (a) finding the continuous compensation D(s) using the ideas in Chapters 1 to 7, and (b) approximating D(s) with difference equations using Tustin’s method or the matched-pole–zero method. In order to analyze a discrete controller design, or any discrete system, the ztransform is used to determine the system’s behavior. The z-transform of a time sequence f (k) is given by Z{f (k)} = F(z) =

∞ 

f (k)z−k

k=0

and has the key property that Z{f (k − 1)} = z−1 F(z).

• •

This property allows us to find the discrete transfer function of a difference equation, which is the digital equivalent of a differential equation for continuous systems. Analysis using z-transforms closely parallels that using Laplace transforms. Normally z-transforms are found using the computer (MATLAB) or looking up in Table 8.1. The discrete Final Value Theorem is lim x(k) = lim (1 − z−1 )X(z),

k→∞

•

z→1

provided that all poles of (1 − z−1 )X(z) are inside the unit circle. For a continuous signal f (t) whose samples are f (k), the poles of F(s) are related to the poles of F(z) by z = esT .

592 Chapter 8 Digital Control •

The following are the most common discrete equivalents: 1. Tustin’s approximation: D(z) = D(s)|

s= T2




z−1 z+1



2. the matched pole–zero approximation: • Map poles and zeros by z = esT . • Add powers of z + 1 to the numerator until numerator and denominator are of equal order or the numerator is one order less than the denominator. • Set the low-frequency gain of D(z) equal to that of D(s). •

•

•

•

•

•

If designing by discrete equivalents, a minimum sample rate of 20 times the bandwidth is recommended. Typically, even faster sampling is useful for best performance. Analog prefilters are commonly placed before the sampler in order to attenuate the effects of high-frequency measurement noise. A sampler aliases all frequencies in the signal that are greater than half the sample frequency to lower frequencies; therefore, prefilter break points should be selected so that no significant frequency content remains above half the sample rate. The discrete model of the continuous plant G(s) preceded by a ZOH is   G(s) −1 . G(z) = (1 − z )Z s The discrete plant model plus the discrete controller can be analyzed using the z-transform or simulated using SIMULINK. Discrete design is an exact design method and avoids the approximations inherent with discrete equivalents. The design procedure entails (a) finding the discrete model of the plant G(s), and (b) using the discrete model to design the compensation directly in its discrete form. The design process is more cumbersome than discrete equivalent design and requires that a sample rate be selected before commencing the design. A practical approach is to commence the design using discrete equivalents, then tune up the result using discrete design. Discrete design using G(z) closely parallels continuous design, but the stability boundary and interpretation of z-plane root locations are different. Figure 8.5 summarizes the response characteristics. If using discrete design, system stability can theoretically be assured when sampling at a rate as slow as twice the bandwidth. However, for good transient performance and random disturbance rejection, best results are obtained by sampling at 10 times the closed-loop bandwidth or faster. In some cases with troublesome vibratory modes, it is sometimes useful to sample more than twice as fast as the vibratory mode.

REVIEW QUESTIONS 1. What is the Nyquist rate? What are its characteristics? 2.

Describe the discrete equivalent design process.

3.

Describe how to arrive at a D(z) if the sample rate is 30 × ωBW .

Problems



593

4.

For a system with a 1 rad/sec bandwidth, describe the consequences of various sample rates.

5.

Give two advantages for selecting a digital processor rather than analog circuitry to implement a controller.

6.

Give two disadvantages for selecting a digital processor rather than analog circuitry to implement a controller.

7.

Describe how to arrive at a D(z) if the sample rate is 5 × ωBW .

PROBLEMS Problems for Section 8.2: Dynamic Analysis of Discrete Systems 8.1

The z-transform of a discrete-time filter h(k) at a 1 Hertz sample rate is H(z) =

1 + (1/2)z−1 . [1 − (1/2)z−1 ][1 + (1/3)z−1 ]

(a) Let u(k) and y(k) be the discrete input and output of this filter. Find a difference equation relating u(k) and y(k). (b) Find the natural frequency and damping coefficient of the filter’s poles. (c) Is the filter stable? 8.2

Use the z-transform to solve the difference equation y(k) − 3y(k − 1) + 2y(k − 2) = 2u(k − 1) − 2u(k − 2), where 

k, 0,

u(k) = y(k) = 0, 8.3

k ≥ 0, k < 0, k < 0.

The one-sided z-transform is defined as F(z) =

∞ 

f (k)z−k .

0

(a) Show that the one-sided transform of f (k + 1) is Z{f (k + 1)} = zF(z) − zf (0). (b) Use the one-sided transform to solve for the transforms of the Fibonacci numbers generated by the difference equation u(k + 2) = u(k + 1) + u(k). Let u(0) = u(1) = 1. [Hint: You will need to find a general expression for the transform of f (k + 2) in terms of the transform of f (k).] (c) Compute the pole locations of the transform of the Fibonacci numbers. (d) Compute the inverse transform of the Fibonacci numbers. (e) Show that, if u(k) represents the kth Fibonacci number, then the ratio u(k + 1)/u(k) √ will approach ( 1+2 5) . This is the golden ratio valued so highly by the Greeks.

8.4

Prove the seven properties of the s-plane-to-z-plane mapping listed in Section 8.2.3.

594 Chapter 8 Digital Control Problems for Section 8.3: Design Using Discrete Equivalents 8.5

A unity feedback system has an open-loop transfer function given by G(s) =

250 . s[(s/10) + 1]

The following lag compensator added in series with the plant yields a phase margin of 50◦ : s/1.25 + 1 . D(s) = 50s + 1 Using the matched pole–zero approximation, determine an equivalent digital realization of this compensator. 8.6

The following transfer function is a lead network designed to add about 60◦ of phase at ω1 = 3 rad/sec: s+1 H(s) = . 0.1s + 1 (a) Assume a sampling period of T = 0.25 sec, and compute and plot in the z-plane the pole and zero locations of the digital implementations of H(s) obtained using (1) Tustin’s method and (2) pole–zero mapping. For each case, compute the amount of phase lead provided by the network at z1 = ejω1 T . (b) Using a log–log scale for the frequency range ω = 0.1 to ω = 100 rad/sec, plot the magnitude Bode plots for each of the equivalent digital systems you found in part (a), and compare with H(s). (Hint: Magnitude Bode plots are given by |H(z)| = |H(ejωT )|.)

8.7

The following transfer function is a lag network designed to introduce a gain attenuation of 10(−20 db) at ω = 3 rad/sec: H(s) =

10s + 1 . 100s + 1

(a) Assume a sampling period of T = 0.25 sec, and compute and plot in the z-plane the pole and zero locations of the digital implementations of H(s) obtained using (1) Tustin’s method and (2) pole–zero mapping. For each case, compute the amount of gain attenuation provided by the network at z1 = ejω1 T . (b) For each of the equivalent digital systems in part (a), plot the Bode magnitude curves over the frequency range ω = 0.01 to 10 rad/sec.

Problem for Section 8.5: Sample Rate Selection 8.8

For the system shown in Fig. 8.20, find values for K, TD , and TI so that the closed-loop poles satisfy ζ > 0.5 and ωn > 1 rad/sec. Discretize the PID controller using (a) Tustin’s method (b) the matched pole–zero method Use MATLAB to simulate the step response of each of these digital implementations for sample times of T = 1, 0.1, and 0.01 sec.

Figure 8.20 Control system for Problem 8.8

R






E

(

K 1  TDs 

1 TIs

)

U

1 s(s  1)

Y

Problems 

595

Problems for Section 8.6: Discrete Design 8.9

Consider the system configuration shown in Fig. 8.21, where G(s) =

40(s + 2) . (s + 10)(s2 − 1.4)

(a) Find the transfer function G(z) for T = 1 assuming the system is preceded by a ZOH. (b) Use MATLAB to draw the root locus of the system with respect to K. (c) What is the range of K for which the closed-loop system is stable? (d) Compare your results of part (c) with the case in which an analog controller is used (that is, where the sampling switch is always closed). Which system has a larger allowable value of K? (e) Use MATLAB to compute the step response of both the continuous and discrete systems with K chosen to yield a damping factor of ζ = 0.5 for the continuous case.

Figure 8.21 Control system for Problem 8.9

R



T




K

G(s)

Y



8.10 Single-Axis Satellite Attitude Control: Satellites often require attitude control for proper orientation of antennas and sensors with respect to Earth. Figure 2.7 shows a communication satellite with a three-axis attitude-control system. To gain insight into the three-axis problem, we often consider one axis at a time. Figure 8.22 depicts this case, where motion is allowed only about an axis perpendicular to the page. The equations of motion of the system are given by I θ¨ = MC + MD , where I = moment of inertia of the satellite about its mass center, MC = control torque applied by the thrusters, MD = disturbance torques,

Figure 8.22 Satellite control schematic for Problem 8.10

u Inertial reference

596 Chapter 8 Digital Control θ = angle of the satellite axis with respect to an inertial reference with no angular acceleration. We normalize the equations of motion by defining MC , I

u=

wd =

MD , I

and obtain θ¨ = u + wd . Taking the Laplace transform yields 1 θ (s) = 2 [u(s) + wd (s)], s which, with no disturbance, becomes 1 θ (s) = 2 = G1 (s). u(s) s In the discrete case in which u is applied through a ZOH, we can use the methods described in this chapter to obtain the discrete transfer function G1 (z) =

T2 θ (z) = u(z) 2



 z+1 . (z − 1)2

(a) Sketch the root locus of this system by hand, assuming proportional control. (b) Draw the root locus using MATLAB to verify the hand sketch. (c) Add discrete velocity feedback to your controller so that the dominant poles correspond to ζ = 0.5 and ωn = 3π/(10T ). (d) What is the feedback gain if T = 1 sec? If T = 2 sec? (e) Plot the closed-loop step response and the associated control time history for T = 1 sec. 8.11 It is possible to suspend a mass of magnetic material by means of an electromagnet whose current is controlled by the position of the mass (Woodson and Melcher, 1968). The schematic of a possible setup is shown in Fig. 8.23, and a photo of a working system at Stanford University is shown in Fig. 9.2. The equations of motion are m¨x = −mg + f (x, I), where the force on the ball due to the electromagnet is given by f (x, I). At equilibrium the magnet force balances the gravity force. Suppose we let I0 represent the current at equilibrium. If we write I = I0 + i, expand f about x = 0 and I = I0 , and neglect higher-order terms, we obtain the linearized equation m¨x = k1 x + k2 i.

(8.46)

Reasonable values for the constants in Eq. (8.46) are m = 0.02 kg, k1 = 20 N/m, and k2 = 0.4 N/A. (a) Compute the transfer function from I to x, and draw the (continuous) root locus for the simple feedback i = −Kx.

Problems

597

Figure 8.23 Schematic of magnetic levitation device for Problems 8.11

(b) Assume that the input is passed through a ZOH, and let the sampling period be 0.02 sec. Compute the transfer function of the equivalent discrete-time plant. (c) Design a digital control for the magnetic levitation device so that the closed-loop system meets the following specifications: tr ≤ 0.1 sec, ts ≤ 0.4 sec, and overshoot ≤ 20%. (d) Plot a root locus with respect to k1 for your design, and discuss the possibility of using your closed-loop system to balance balls of various masses. (e) Plot the step response of your design to an initial disturbance displacement on the ball, and show both x and the control current i. If the sensor can measure x only over a range of ±1/4 cm and the amplifier can provide a current of only 1 A, what is the maximum displacement possible for control, neglecting the nonlinear terms in f (x, I)? 8.12 Repeat Problem 5.27 in Chapter 5 by constructing discrete root loci and performing the designs directly in the z-plane. Assume that the output y is sampled, the input u is passed through a ZOH as it enters the plant, and the sample rate is 15 Hz. 8.13 Design a digital controller for the antenna servo system shown in Figs. 3.61 and 3.62 and described in Problem 3.31. The design should provide a step response with an overshoot of less than 10% and a rise time of less than 80 sec. (a) What should the sample rate be? (b) Use discrete equivalent design with the matched pole–zero method. (c) Use discrete design and the z-plane root locus. 8.14 The system G(s) =

1 (s + 0.1)(s + 3)

is to be controlled with a digital controller having a sampling period of T = 0.1 sec. Using a z-plane root locus, design compensation that will respond to a step with a rise time tr ≤ 1 sec and an overshoot Mp ≤ 5%. What can be done to reduce the steady-state error?

598 Chapter 8 Digital Control 8.15 The transfer function for pure derivative control is D(z) = KTD

z−1 , Tz

where the pole at z = 0 adds some destabilizing phase lag. Can this phase lag be removed by using derivative control of the form D(z) = KTD

(z − 1) ? T

Support your answer with the difference equation that would be required and discuss the requirements to implement it.

9

Nonlinear Systems

Perspective on Nonlinear Systems All systems are nonlinear, especially if large signals are considered. On the other hand, almost all physical systems can be well approximated by linear models if the signals are small. For example, if θ is small, then sin(θ ) ≈ θ and cos(θ ) ≈ 1. Similarly, in analog electronic devices such as amplifiers, the operation will be nearly linear if the signals are small with respect to the supply voltage. Finally, as we will consider later in this chapter in an optional section, Lyapunov showed that if the linear approximation of a system is stable near an equilibrium point, then the truly nonlinear system will be stable for some neighborhood of the equilibrium point. For all these reasons, the analysis and design methods presented thus far in this book have considered only the enormously powerful techniques available for linear models. However, if the signals cause a device to saturate or if the system includes nonlinearities that are active for small signals, such as some kinds of friction, then the nonlinear effects must be taken into account to explain the behavior of the system. In this chapter, a few of the tools available for this purpose will be described.

599

600 Chapter 9 Nonlinear Systems

Chapter Overview Because every nonlinear system is in many ways unique, a vast number of approaches are used in nonlinear control design. The approaches to analysis and design of nonlinear systems that we will describe may be classified under four categories. In Section 9.2 methods of reducing the problem to a linear model are discussed. In most cases, considering the small signal approximation is adequate. In some cases there are nonlinearities for which an inverse can be found, and placing the inverse before the physical nonlinearity results in an overall system that responds linearly. In yet other cases, some nonlinear models can be reduced to an exact linear form by the clever use of feedback, in a technique called “computed torque” in the field of robotics. The second category is a heuristic approach based on considering the nonlinearity to be a varying gain. In Section 9.3 cases are considered for which the nonlinearity has no memory as, for example, with an amplifier whose output saturates when the signal gets large. The idea is to consider the amplifier as if its gain begins to be reduced as the signal gets large. Because the root locus is based on evaluating the system characteristic roots as gain changes, this point of view leads to a heuristic use of the root locus to predict how such a system will respond to changing input signal sizes. Section 9.4 treats cases in which the nonlinearity has dynamics or memory; then the root locus is not useful. For these cases a technique introduced by Kochenburger in 1950 known as the describing function can be used. To apply this method, a sinusoid is applied to the nonlinear part of the system and the first harmonic of the periodic response is computed. The ratio of the input to the output is taken as if it were a linear but variable frequency response. Thus the Nyquist plot is the natural domain in which to consider the system behavior. While the heuristic approaches may give very useful insight into the system’s behavior, they cannot be used to decide if the system is guaranteed to be stable. For this, one must turn to the analysis of stability as studied in control theory. The most famous of these theories is that of internal stability developed by Lyapunov. As an introduction to the idea of a system response as a trajectory in space, Section 9.5 describes analysis in the phase plane and then presents the stability theory. Examples are given of using the stability theorem to guide design of a controller so the system is guaranteed to be stable if the initial assumptions about the system hold. With these methods, the control engineer is given a start on the path to the effective understanding and design of real control problems. Finally Section 9.6 provides a historical perspective for the material in this chapter.

9.1 Introduction and Motivation: Why Study Nonlinear Systems? It is intuitively clear that at some level of signal strength any physical system will be nonlinear and that some systems are nonlinear at any and all signal levels. On the other hand, we began our study by developing linear approximate models, and all our design methods thus far have been based on the assumption that the plant can

9.1 Introduction and Motivation: Why Study Nonlinear Systems?

601

be represented by a linear transfer function. In this chapter we shall give some of the reasons for believing that all the time spent studying linear techniques was not a waste of time, but we shall also try to explain why it is very important to understand how to take nonlinear effects into account in control system design. We begin by showing that we can combine the root-locus technique, which plots roots of the characteristic equation as a function of various gain values, with the observation that many nonlinear elements can be viewed as a gain that changes as signal level changes. While the method is, at this point, entirely heuristic, the simulation results are very promising. Many properties of systems containing such zero-memory nonlinear elements can be predicted by plotting a root locus versus gain at the point of the nonlinearity. However, the method, as presented, is not on a firm foundation, and the designer is left to wonder if there is some unexplored region of the real state space or the signal spaces where catastrophe awaits. After all, the model is an approximation, and no matter how extensive the simulation, it is not possible to cover every situation. Following the use of the root locus, we turn to methods based on the frequency response. One of the great advantages of the frequency response is that in many cases one can obtain the transfer function experimentally on the real system. In the most basic approach, a sinusoidal signal is applied to the system and the amplitude and phase of the output sinusoid are measured. However, noise and inevitable nonlinear effects cause the output to be more complicated than a simple sinusoid, so the designer extracts the fundamental component and treats it as if it is the whole story. One gets the same result if a spectrum analyzer is used to compute a transfer function. What has been done is to compute what Kochenburger called the describing function. From this point of view, a describing function can be defined for nonlinear elements, including those with memory. Again, simulations are promising and many useful designs are done with this technique but, as with the use of root locus to design nonlinear systems, this method is also on shaky ground.1 So what is to be made of this situation? The only possibility is to face up to the facts and take on nonlinear behavior directly. Fortunately, a firm foundation in mathematics was established whenA. M. Lyapunov published his work on the stability of motion in 1892. This work was translated into French in 1907 and recovered in a control context by Kalman and Bertram in 1960. Lyapunov gave two methods for the study of stability. For his first method, he considered stability based on the linear approximation, the very thing required to justify our concentration on that approach in this book. He proved the remarkable result that if the linear approximation is strictly stable, having all roots in the left half-plane (LHP), then the nonlinear system will have a region of stability around the equilibrium point where the linear approximation applies. Furthermore, he proved that if the linear approximation has at least one root in the right half-plane (RHP), then the nonlinear system cannot have any region of stability in the neighborhood of the equilibrium. The size of the stable region in the state space is not given by the linear terms but is included in the construction used for the proof. That construction constituted his second method. Lyapunov’s second method is based on the mathematical equivalent of finding a scalar function that

1And as we live in California, we know how dangerous it is to be on shaky ground.

602 Chapter 9 Nonlinear Systems describes the internal energy stored in the system. He proved that if such a function is constructed, and if the derivative of the function is negative on trajectories of the equations of motion, then the function and the state on which it depends will eventually drain away and the state will come to rest at the equilibrium point. A function having these properties is called a Lyapunov function. Of course, this simple description omits a great deal of complexity; for example, there are dozens of definitions of stability. However, the concept remains that if a Lyapunov function can be found, then the system on which it is based will be stable. As described, the theory gives a sufficient condition for stability. If a Lyapunov function is not found, the designer does not know if one does not exist or if the search has just been inadequate. A great deal of research has been directed toward finding Lyapunov functions for particular classes of nonlinear systems. Lyapunov’s methods are based on differential equations in normal or state form and are thus concerned with internal stability. Frequency-response methods, on the other hand, are external measures, and there has been interest in developing stability results based on the external response of the system. One such method is the circle criterion, which we will describe in this chapter also. The method can be described as considering the energy seen at the terminal of the system and noting if it is always flowing into the terminals. If so, it is reasonable to assume that eventually all energy will be gone and the system will be stable. For a formal proof of the method, researchers have turned to Lyapunov’s second method, but the result is expressed in terms of external properties, such as the Nyquist plot of the linear portion of the system that faces the nonlinear elements. Again, this tool gives a basis for setting a firm foundation under a method of design for a particular class of nonlinear systems. As should be clear at this point, the theory of nonlinear control is a vast and sophisticated topic, and in this book we can give only a brief introduction to a small part of it. However, the foundation of control design rests on this theory, and the more the designer understands of the theory, the better he or she understands both the limits and the opportunities of problems. It is our hope that by considering this material, students will be stimulated to further their profitable study of this fascinating topic.

9.2 Analysis by Linearization Three methods of reducing some nonlinear systems to a suitable linear model are presented in this section. The differential equations of motion for almost all processes selected for control are nonlinear. On the other hand, both analysis and control design methods we have discussed so far are much easier for linear than for nonlinear models. Linearization is the process of finding a linear model that approximates a nonlinear one. Fortunately, as Lyapunov proved over 100 years ago, if a small-signal linear model is valid near an equilibrium and is stable, then there is a region (which may be small, of course) containing the equilibrium within which the nonlinear system is stable. So we can safely make a linear model and design a linear control for it such that, at least in the neighborhood of the equilibrium, our design will be stable. Since a very important role of feedback control is to maintain the process variables near equilibrium, such small-signal linear models are a frequent starting point for control design.

9.2 Analysis by Linearization

603

An alternative approach to obtain a linear model for use as the basis of control system design is to use part of the control effort to cancel the nonlinear terms and to design the remainder of the control based on linear theory. This approach—linearization by feedback—is popular in the field of robotics, where it is called the method of computed torque. It is also a research topic for control of aircraft. Section 9.2.2 takes a brief look at this method. Finally, some nonlinear functions are such that an inverse nonlinearity can be found to be placed in series with the nonlinearity so that the combination is linear. This method is often used to correct mild nonlinear characteristics of sensors or actuators that have small variations in use, as discussed in Section 9.2.3.

9.2.1 Linearization by Small-Signal Analysis For a system with smooth nonlinearities and a continuous derivative, one can compute a linear model that is valid for small signals. In many cases these models can be used for design. A nonlinear differential equation is an equation for which the derivatives of the state have a nonlinear relationship to the state itself and/or the control. In other words, the differential equations cannot be written in the form2 x˙ = Fx + Gu, but must be left in the form x˙ = f(x, u). (9.1) For small-signal linearization we first determine equilibrium values of xo , uo , such that x˙ o = 0 = f(xo , uo ), and we let x = xo + δx and u = uo + δu. We then expand the nonlinear equation in terms of the perturbations from these equilibrium values, which yields x˙ o + δ x˙ ∼ = f(xo , uo ) + Fδx + Gδu, where F and G are the best linear fits to the nonlinear function f(x, u) at xo and uo , computed as     ∂f ∂f F= and G = . (9.2) ∂x xo ,uo ∂u xo ,uo Subtracting out the equilibrium solution, this reduces to δ x˙ = Fδx + Gδu,

(9.3)

which is a linear differential equation approximating the dynamics of the motion about the equilibrium point. Normally, the δ notation is dropped and it is understood that x and u refer to the deviation from the equilibrium. In developing the models discussed so far in this book, we have encountered nonlinear equations on several occasions: the pendulum in Example 2.5, the hanging crane in Example 2.7, the AC induction motor in Section 2.3, the tank flow in Example 2.16, and the hydraulic actuator in Example 2.17. In each case, we assumed either that the motion was small or that motion from some operating point was small, so that nonlinear functions were approximated by linear functions. The steps followed in those examples essentially involved finding F and G in order to linearize the differential equations to the form of Eq. (9.3), as illustrated in the next several examples. The linearization functions in MATLAB
include linmod and linmod2. 2 This equation assumes the system is time invariant. A more general expression would be x˙ = f(x,u, t).

604 Chapter 9 Nonlinear Systems

EXAMPLE 9.1

Linearization of Nonlinear Pendulum Consider the nonlinear equations of motion of the simple pendulum in Example 2.5. Derive the equilibrium points for the system and determine the corresponding smallsignal linear models. Solution. The equation of motion is g Tc θ¨ + sin θ = . (9.4)  m2 We can rewrite the equation of motion in state-variable form, with x = [ x1 x2 ]T = [ θ θ˙ ]T , as     x2 f1 (x, u) = = f(x, u), x˙ = f2 (x, u) −ωo2 sin x1 + u  Tc g where ωo = and u = . To determine the equilibrium state, suppose that the  m2 (normalized) input torque has a nominal value of uo = 0. Then x˙ 1 = θ˙ = 0, g x˙ 2 = θ¨ = − sin θ = 0,  so that the equilibrium conditions correspond to θo = 0, π (i.e., the downward and the inverted pendulum at rest configurations, respectively). The equilibrium state and the input are xo = [ θo 0 ]T , uo = 0, and the state-space matrices are given by ⎡ ⎤ ∂f1 ∂f1   ⎢ ∂x1 ∂x2 ⎥ 0 1 ⎢ ⎥ = F=⎣ , −ωo2 cos θo 0 ∂f2 ∂f2 ⎦ ∂x1 ∂x2 xo ,uo ⎤ ⎡ ∂f1   ⎢ ∂u ⎥ 0 ⎥ ⎢ = . G=⎣ 1 ∂f2 ⎦ ∂u xo ,uo The linear system has eigenvalues of ±jωo and ±ωo corresponding to θo = 0 and π , respectively, with the latter inverted case being unstable as expected.

EXAMPLE 9.2

Linearization of Motion in a Ball Levitator Figure 9.1 shows a magnetic bearing used in large turbo machinery. The magnetics are energized using feedback control methods so that the axle is always in the center and never touches the magnets, thus keeping friction to an almost nonexistent level. A simplified version of a magnetic bearing that can be built in a laboratory is shown in Fig. 9.2, where one electromagnet is used to levitate a metal ball. The physical arrangement of the levitator is depicted in Fig. 9.3. The equation of motion of the ball, derived from Newton’s law, Eq. (2.1), is m¨x = fm (x, i) − mg,

(9.5)

9.2 Analysis by Linearization

605

Figure 9.1 A magnetic bearing Source: Photo courtesy of Magnetic Bearings, Inc.

Figure 9.2 Magnetic ball levitator used in the laboratory Source: Photo courtesy of Gene Franklin

where the force fm (x, i) is caused by the field of the electromagnet. Theoretically, the force from an electromagnet falls off with an inverse square relationship to the distance from the magnet, but the exact relationship for the laboratory levitator is difficult to derive from physical principles because its magnetic field is so complex. However, the forces can be measured with a scale. Figure 9.4 shows the experimental curves for a ball with a 1-cm diameter and a mass of 8.4 × 10−3 kg. At the value for the current of i2 = 600 mA and the displacement x1 shown in the figure, the magnetic force fm just cancels the gravity force, mg = 82 × 10−3 N. (The mass of the ball is 8.4 × 10−3 kg, and the acceleration of gravity is 9.8 m/sec2 .) Therefore the point (x1 , i2 ) represents an equilibrium. Using the data, find the linearized equations of motion about the equilibrium point.

606 Chapter 9 Nonlinear Systems Figure 9.3 Model for ball levitation

i

x

Figure 9.4 Experimentally determined force curves

200 i1  700 mA Force, fm (103 N)

160 i2  600 mA 120

i3  500 mA

80 40 0

Slope Kx

1

2 x1 3 4 Distance, x (mm)

5

6

Solution. First we write, in expanded form, the force in terms of deviations from the equilibrium values x1 and i2 : (9.6) fm (x1 + δx, i2 + δi) ∼ = fm (x1 , i2 ) + Kx δx + Ki δi. The linear gains are found as follows: Kx is the slope of the force versus x along the curve i = i2 , as shown in Fig. 9.4, and is found to be about 14 N/m. Ki is the change of force with current for the value of fixed x = x1 . We find that for i = i1 = 700 mA at x = x1 , the force is about 122 × 10−3 N, and at i = i3 = 500 mA at x = x1 , it is about 42 × 10−3 N. Thus 80 × 10−3 N 122 × 10−3 − 42 × 10−3 Ki ∼ = = 700 − 500 200 mA −3 ∼ = 400 × 10 N/A ∼ = 0.4 N/A. Substituting these values into Eq. (9.6) leads to the following linear approximation for the force in the neighborhood of equilibrium: ∼ 82 × 10−3 + 14δx + 0.4δi. fm =

9.2 Analysis by Linearization

607

Substituting this expression into Eq. (9.5) and using the numerical values for mass and gravity force, we get for the linearized model (8.4 × 10−3 )¨x = 82 × 10−3 + 14δx + 0.4δi − 82 × 10−3 . Because x = x1 + δx, then x¨ = δ x¨ . The equation in terms of δx is thus (8.4 × 10−3 )δ x¨ = 14δx + 0.4δi, δ x¨ = 1667δx + 47.6δi,

(9.7)

which is the desired linearized equation of motion about the equilibrium point. A logical state vector is x = [δx δ x˙ ]T , which leads to the standard matrices     0 1 0 F= and G = 1667 0 47.6 and the control u = δi.

EXAMPLE 9.3

Linearization of the Water Tank Revisited Repeat the linearization of Example 2.16 using the concepts presented in this section. Solution. Equation (2.75) may be written as x˙ = f (x, u), (9.8) 1 √ 1 1 √ where x  h, u  win , and f (x, u) = − RAρ p1 − pa + Aρ win = − RAρ ρgh − pa + 1 Aρ win .

The linearized equations are of the form δ x˙ = Fδx + Gδu,

where [F]xo ,uo =

(9.9)

    ∂f ∂f ∂ 1  = ρgh − pa = − ∂x ∂h ho ,uo ∂h RAρ ho ,uo

=−

1 1 g g =− √ √ 2AR ρgho − pa 2AR po − pa

(9.10) (9.11)

and

∂f ∂f 1 = . (9.12) = ∂u ∂win Aρ However, note that some flow is required to maintain the system in equilibrium so that Eq. (9.9) is valid; specifically, we see from Eq. (2.75) that [G]xo ,uo =

1√ po − pa for h˙ = 0 (9.13) R and the δu in Eq. (9.9) is δwin , where win = wino +δwin . Therefore, Eq. (9.9) becomes uo = wino =

δ h˙ = Fδh + Gδwin = Fδh + Gwin − G and matches Eq. (2.78) precisely.

1√ po − pa R

(9.14)

608 Chapter 9 Nonlinear Systems 9.2.2 Linearization by Nonlinear Feedback Nonlinear feedback

Linearization by feedback is accomplished by subtracting the nonlinear terms out of the equations of motion and adding them to the control. The result is a linear system, provided that the computer implementing the control has enough capability to compute the nonlinear terms fast enough and the resulting control does not cause the actuator to saturate. A more detailed understanding of the method is best achieved through example.

EXAMPLE 9.4

Linearization of the Nonlinear Pendulum Consider the equation of a simple pendulum developed in Example 2.5 [Eq. (2.21)]. Linearize the system by using nonlinear feedback. Solution. The equation of motion is ml 2 θ¨ + mgl sin θ = Tc .

(9.15)

If we compute the torque to be Tc = mgl sin θ + u,

(9.16)

ml2 θ¨ = u.

(9.17)

then the motion is described by

Eq. (9.17) is a linear equation no matter how large the angle θ becomes. We use it as the model for purposes of control design because it enables us to use linear analysis techniques. The resulting linear control will provide the value of u based on measurements of θ ; however, the value of the torque actually sent to the equipment would derive from Eq. (9.16). For robots with two or three rigid links, this computedtorque approach has led to effective control. It is also being researched for the control of aircraft, where the linear models change considerably in character with the flight regime.

9.2.3 Linearization by Inverse Nonlinearity Inverse nonlinearity

The simplest case of the introduction of nonlinearities into a control design is that of inverse nonlinearities. It is sometimes possible to reverse the effect of some nonlinearities. For example, suppose we have a system whose output is the square of the signal of interest: y = x2 .

(9.18)

One clever and rather obvious technique is to undo the nonlinearity by preceding the physical nonlinearity with a square root nonlinearity,  x = (.), (9.19) as shown in the next example. The overall cascaded system would then be linear.

9.3 Equivalent Gain Analysis Using the Root Locus

609

Controller Lamp R






V'

D(s)



0



V

P  V2

RTP plant P

G(s) 

0.5 s  0.08

Y

Figure 9.5 Linearization through inverse nonlinearity

EXAMPLE 9.5

Linearization of the Rapid Thermal Processing (RTP) System Consider the RTP system that uses a nonlinear lamp as an actuator, as shown in Fig. 9.5. Suppose the input to the lamp is voltage V and the output is power P and they are related by P = V 2. Design an inverse nonlinearity to linearize the system. Solution. We simply precede the lamp input nonlinearity with a square root nonlinearity √ V = V
. The overall open-loop cascaded system is now linear for any value of the voltage: Y = G(s)P = G(s)V 2 = G(s)V
. Thus we can use linear control design techniques for the dynamic compensator, D(s). Note that a nonlinear element has been inserted in front of the square root element to ensure that the input to this block remains nonnegative at all times. The controller is then implemented as shown in Fig. 9.5. For a detailed application of this method for control design, we refer the reader to the RTP case study in Section 10.6.

9.3 Equivalent Gain Analysis Using the Root Locus

Memoryless nonlinearity

As we have tried to make clear, every real control system is nonlinear, and the linear analysis and design methods we have described so far use linear approximations to the real models. There is one important category of nonlinear systems for which linearization is not appropriate and for which some significant analysis (and design) can be done. This category comprises the systems in which the nonlinearity has no dynamics and is well approximated as a gain that varies as the size of its input signal varies. Sketches of a few such nonlinear system elements and their common names are shown in Fig. 9.6. The stability of systems with memoryless nonlinearities can be studied heuristically using the root locus. The technique is to replace the memoryless nonlinearity by an equivalent gain K, and a root locus is plotted versus this gain. For a range of input signal amplitudes, the equivalent gain will take on a range of values, and the

610 Chapter 9 Nonlinear Systems Figure 9.6 Nonlinear elements with no dynamics: (a) saturation; (b) relay; (c) relay with dead zone; (d) gain with dead zone; (e) preloaded spring, or coulomb plus viscous friction; (f) quantization

y

y

N

h

N

h

u

h N

N

u

h N

N

(a)

(b)

y Slope K0

Slope K0

q u

u

h

2q

N

h

q

N

(d)

u

(c)

y

y

Figure 9.7 Dynamic system with saturation

y

u

2q

(e)

(f)

Saturation 0.4

r








K

0.4

0.4

s1 s2

y

0.4

closed-loop roots of the system are examined in this range as if the gain were fixed. This is illustrated by the next several examples.

EXAMPLE 9.6

Changing Overshoot and Saturation Nonlinearity Consider the system with saturation shown in Fig. 9.7. Determine the stability properties of the system using the root locus technique. Solution. The root locus of this system versus K with the saturation removed is given by Fig. 9.8. At K = 1 the damping ratio is ζ = 0.5. As the gain is reduced, the locus shows that the roots move toward the origin of the s-plane with less and less damping. Plots of the step responses of this system were obtained using the SIMULINK
program. A series of step inputs r with magnitudes r0 = 2, 4, 6, 8, 10, and 12 was introduced to the system, and the results are shown in Fig. 9.9. As long as the signal entering the saturation remains less than 0.4, the system will be linear and should behave according to the roots at ζ = 0.5. However, notice that as the input gets larger, the response has more and more overshoot and slower and

9.3 Equivalent Gain Analysis Using the Root Locus Figure 9.8 Root locus of (s+1)/s2 , the system in Fig. 9.7 with the saturation removed

Im(s)

K1

1

Figure 9.9 Step responses of system in Fig. 9.7 for various input step sizes

611

Re(s)

20 18

12

16

10

14 12 8

y 10 6

8 6

4

4

r2

2 0

5

10

15 20 Time (sec)

25

30

K

N/a Gain

Figure 9.10 General shape of the effective gain of saturation

0

a

Input magnitude

u

slower recovery. This can be explained by noting that larger and larger input signals correspond to smaller and smaller effective gain K, as seen in Fig. 9.10. From the root-locus plot of Fig. 9.8, we see that as K decreases, the closed-loop poles move closer to the origin and have a smaller damping ζ . This results in the longer rise and settling times, increased overshoot, and greater oscillatory response.

EXAMPLE 9.7

Stability of Conditionally Stable System Using the Root Locus

A nonlinear example: stability depends on input magnitude

As a second example of a nonlinear response described by signal-dependent gain, consider the system with a saturation nonlinearity as shown in Fig. 9.11. Determine whether the system is stable.

612 Chapter 9 Nonlinear Systems

Figure 9.12 Root locus for G(s) = (s + 1)2 /s3 from system in Fig. 9.11

1 

R




u

K

(s  1)2 s3

1



Y

3 2 Imaginary axis

Figure 9.11 Block diagram of a conditionally stable system

1 0 1 2 3 6

Amplitude

Figure 9.13 Step responses of system in Fig. 9.11

Conditional stability

9 8 7 6 5 4 3 2 1 0 1

5

4

3

2 1 Real axis

0

1

2

r  3.475 3 2 1

0

2

4

6

8 10 12 14 16 18 20 Time (sec)

Solution. The root locus for the system, excluding the saturation, is plotted in Fig. 9.12. From this locus we can readily calculate that the imaginary axis crossing occurs at ω0 = 1 and K = 21 . Systems such as this, which are stable for (relatively) large gains but unstable for smaller gains, are called conditionally stable systems. If K = 2, which corresponds to ζ = 0.5 on the locus, the system would be expected to show responses consistent with ζ = 0.5 for small reference input signals. However, as the reference input size gets larger, the equivalent gain would get smaller due to the saturation, and the system would be expected to become less well damped. Finally, the system would be expected to become unstable at some point for large inputs. Step responses from nonlinear simulation of the system with K = 2 for input steps of size

9.3 Equivalent Gain Analysis Using the Root Locus

613

r0 = 1.0, 2.0, 3.0, and 3.4 are shown in Fig. 9.9. These responses confirm our predictions. Furthermore, the marginally stable case shows oscillations near 1 rad/sec, which is predicted by the frequency at the point at which the root locus crosses into the RHP.

EXAMPLE 9.8

Analysis and Design of System with Limit Cycle Using the Root Locus

A nonlinear example: an oscillatory system with saturation

The final illustration of the use of the root locus to give a qualitative description of the response of a nonlinear system is based on the block diagram in Fig. 9.14. Determine whether the system is stable and find the amplitude and the frequency of the limitcycle. Modify the controller design to minimize the effect of limit-cycle oscillations. Solution. This system is typical of electromechanical control problems in which the designer perhaps at first is not aware of the resonant mode corresponding to the denominator term s2 + 0.2s + 1, (ω = 1, ζ = 0.1). The root locus for this system versus K, excluding the saturation, is sketched in Fig. 9.15. The imaginary-axis crossing can be verified to be at ω0 = 1, K = 0.2; thus a gain of K = 0.5 is enough to force the roots of the resonant mode into the RHP, as shown by the dots. If the system gain is set at K = 0.5, our analysis predicts a system that is initially unstable but becomes stable as the gain decreases. Thus we would expect the response of the system with the saturation to build up due to the instability until the magnitude is sufficiently large that the effective gain is lowered to K = 0.2 and then stop growing! Plots of the step responses with K = 0.5 for three steps of size r0 = 1, 4, and 8 are shown in Fig. 9.16, and again our heuristic analysis is exactly correct: The error builds up to a fixed amplitude and then starts to oscillate at a fixed amplitude. The oscillations have a frequency of ≈1 rad/sec and hold constant amplitude regardless of the step sizes of the input. In this case, the response always approaches a periodic solution of fixed amplitude known as a limit cycle, so-called because the response is cyclic and is approached in the limit as time grows large. We can return to Fig. 9.13 and be easily convinced that the first transient to a step of size 3 is nearly a sinusoid. We can predict that the system is just on the border of stability for an equivalent gain corresponding to a root locus gain of 1/2 when the locus crosses into the RHP. In order to prevent the limit cycle, the locus has to be modified by compensation so that no branches cross into the RHP. One common method of doing this for a lightly damped oscillatory mode is to place compensation zeros near the poles at a frequency such that the angle of departure of the root-locus branch from these poles is toward the LHP, a procedure called phase stabilization earlier. Example 5.8 for collocated mechanical motion demonstrated that a pole–zero

Limit cycle

Figure 9.14 Block diagram of a system with an oscillatory mode

0.1 R








K

0.1

0.1

U

1 s2  0.2s  1

1 s

Y

614 Chapter 9 Nonlinear Systems Figure 9.15 Root locus for the system in Fig. 9.14

1.5

Imaginary axis

1

K  0.5

0.5 0 0.5 1 1.5 3 2.5 2 1.5 1 0.5 Real axis

Figure 9.16 Step responses of system in Fig. 9.14

9

0

0.5

1

r8

8 7 Amplitude

6 r4

5 4 3

r1

2 1 0

0

50

100

150

Time (sec)

Figure 9.17 Root locus including compensation

Im(s)

0.5

10.0

1.0 0.5

Re(s)

0.5

pair located in this manner will often cause a locus branch to go from the pole to the zero, looping to the left, and thus staying away from the RHP. Figure 9.17 shows the root locus for the system, 1/[s(s2 + 0.2s + 1.0)], including a notch compensation with zeros located as just discussed. In addition, the compensation also includes two poles to make the compensation physically realizable. In this case, both poles were

9.3 Equivalent Gain Analysis Using the Root Locus Figure 9.18 Block diagram of the system with a notch filter

615

0.1 

R






u

1 2

100 0.81

Figure 9.19 Step responses of the system in Fig. 9.18

1 s2  0.2s  1

0.1

1 s

Y

s2  0.18s  0.81 (s  10)2

4.5 4

r4

Amplitude

3.5 3 2.5

r2

2 1.5 1 0.5 0

0

5

10 15 20 25 30 35 40 45 50 Time (sec)

placed at s = −10, fast enough to not cause stability problems with the system, yet slow enough that high-frequency noise would not be amplified too much. Thus the compensation used for the root locus is s2 + 0.18s + 0.81 , (s + 10)2 where the gain of 123 has been selected to make the compensation’s DC gain equal to unity. This notch filter compensation attenuates inputs in the vicinity of ωn2 = 0.81 or ωn = 0.9 rad/sec, so that any input from the plant resonance is attenuated and is thus prevented from detracting from the stability of the system. Figure 9.18 shows the system, including the notch filter, and Fig. 9.19 shows the time response for two step inputs. Both inputs, r0 = 2 and 4, are sufficiently large that the nonlinearity is saturated initially; however, because the system is unconditionally stable, the saturation results only in lowering the gain, so the response is slower than predicted by linear analysis but still stable, as also predicted by our piecewise linear analysis. In both cases the nonlinearity eventually becomes unsaturated, and the system stabilizes to its new commanded value of r. D(s) = 123

9.3.1 Integrator Antiwindup In any control system the output of the actuator can saturate because the dynamic range of all real actuators is limited. For example, a valve saturates when it is fully

616 Chapter 9 Nonlinear Systems open or closed, the control surfaces on an aircraft cannot be deflected beyond certain angles from their nominal positions, electronic amplifiers can produce only finite voltage outputs, etc. Whenever actuator saturation happens, the control signal to the process stops changing and the feedback path is effectively opened. If the error signal continues to be applied to the integrator input under these conditions, the integrator output will grow (windup) until the sign of the error changes and the integration turns around. The result can be a very large overshoot, as the output must grow to produce the necessary unwinding error, and poor transient response is the result. In effect, the integrator is an unstable element in open loop and must be stabilized when saturation occurs.3 Consider the feedback system shown in Fig. 9.20. Suppose a given reference step is more than large enough to cause the actuator to saturate at umax . The integrator continues integrating the error e, and the signal uc keeps growing. However, the input to the plant is stuck at its maximum value, namely u = umax , so the error remains large until the plant output exceeds the reference and the error changes sign. The increase in uc is not helpful since the input to the plant is not changing, but uc may become quite large if saturation lasts a long time. It will then take a considerable negative error e and the resulting poor transient response to bring the integrator output back to within the linear band where the control is not saturated. The solution to this problem is an integrator antiwindup circuit, which “turns off” the integral action when the actuator saturates. (This can be done quite easily with logic, if the controller is implemented digitally, by including a statement such as “if |u| = umax , kI = 0;” see Chapter 8.) Two equivalent antiwindup schemes are shown in Fig. 9.21(a, b) for a PI controller. The method in Fig. 9.21(a) is somewhat easier to understand, whereas the one in Fig. 9.21(b) is easier to implement, as it does not require a separate nonlinearity but uses the saturation itself.4 In these schemes, as soon as the actuator saturates, the feedback loop around the integrator becomes active and acts to keep the input to the integrator at e1 small. During this time the integrator Figure 9.20 Feedback system with actuator saturation

kP

R






e e1

kI s

u  




uc

umin

u umax uc

Plant

Y

3 In process control, integral control is usually called reset control, and so integrator windup is often called

reset windup. Without integral control, a given setpoint of, say, 10 results in a response of less value, say, 9.9. The operator must then reset to 10.1 to bring the output to the desired value of 10. With integral control the controller automatically brings the output to 10 with a setpoint of 10; hence the integrator does automatic reset. 4 In some cases, especially with mechanical actuators such as an aircraft control surface or a flow control

valve, it is not desirable and may cause damage to have the physical device bang against its stops. In such cases it is common practice to include an electronic saturation with lower limits than those of the physical device, so that the system hits the electrical stops just before the physical device will saturate.

9.3 Equivalent Gain Analysis Using the Root Locus Figure 9.21 Integrator antiwindup techniques

kP 

e

u





uI

kI s




617



umin

uc




u

umax uc

u Ka umax uc

(a) kP e






u 

uI

kI s






umin

uc

umax uc

u






Ka



(b) kP e






kI s

 




uc 






Ka



umax

(c)

e

kPs  kI s  KakI

uc

(d)

essentially becomes a fast first-order lag. To see this, note that we can redraw the portion of the block diagram in Fig. 9.21(a) from e to uc as shown in Fig. 9.21(c). The integrator part then becomes the first-order lag shown in Fig. 9.21(d). The antiwindup gain, Ka , should be chosen to be large enough that the antiwindup circuit keeps the input to the integrator small under all error conditions. The effect of the antiwindup is to reduce both the overshoot and the control effort in the feedback system. Implementation of such antiwindup schemes is a necessity in any practical application of integral control, and omission of this technique may lead to serious deterioration of the response. From the point of view of stability, the effect of the saturation is to open the feedback loop and leave the open loop plant

618 Chapter 9 Nonlinear Systems kp 2

Control

kI  Step








1 s

4 





Plant 1 s




Ka

Output

Saturation

Integrator 

10






Figure 9.22 SIMULINK
diagram for the antiwindup example

with a constant input and the controller as an open-loop system with the system error as input. Purpose of antiwindup

EXAMPLE 9.9

The purpose of the antiwindup is to provide local feedback to make the controller stable alone when the main loop is opened by signal saturation, and any circuit that does this will perform as antiwindup.5

Antiwindup Compensation for a PI Controller Consider a plant with the transfer function for small signals, G(s) =

1 , s

and a PI controller, 4 kI =2+ , s s in the unity feedback configuration. The input to the plant is limited to ±1.0. Study the effect of antiwindup on the response of the system. Dc (s) = kp +

Solution. Suppose we use an antiwindup circuit with a feedback gain of Ka = 10, as shown in the SIMULINK
block diagram of Fig. 9.22. Figure 9.23(a) shows the step response of the system with and without the antiwindup element. Figure 9.23(b) shows the corresponding control effort. Note that the system with antiwindup has substantially less overshoot and less control effort.

5A more sophisticated scheme might use an antiwindup feedback at a lower level of saturation than that

imposed by the actuator, so PD control continues for a time after integration has been stopped. Any such scheme needs to be analyzed carefully to evaluate its performance and to assure stability.

619

9.4 Equivalent Gain Analysis Using Frequency Response: Describing Functions 1.6 Without antiwindup 1.4 With antiwindup 1.2 Control

Output

1.0 0.8 0.6 0.4 0.2 0

0

1

2

3

4 5 6 Time (sec)

7

8

9

10

(a)

1.2 1.0 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1.0 1.2

Without antiwindup With antiwindup

0

1

2

3

4 5 6 Time (sec)

7

8

9

10

(b)

Figure 9.23 Integrator antiwindup: (a) step response; (b) control effort

9.4 Equivalent Gain Analysis Using Frequency Response: Describing Functions The behavior of systems containing any one of the nonlinearities shown in Fig. 9.6 can be qualitatively described by considering the nonlinear element as a varying signaldependent gain. For example, with the saturation element (Fig. 9.6a), it is clear that for input signals with magnitudes of less than h, the nonlinearity is linear with the gain N/h. However, for signals larger than h, the output size is bounded by N, while the input size can get much larger than h, so once the input exceeds h, the ratio of output to input goes down. Thus, saturation has the gain characteristics shown in Fig. 9.10. All actuators saturate at some level. If they did not, their output would increase to infinity, which is physically impossible. An important aspect of control system design is sizing the actuator, which means picking the size, weight, power required, cost, and saturation level of the device. Generally, higher saturation levels require bigger, heavier, and more costly actuators. From the control point of view, the key factor that enters into the sizing is the effect of the saturation on the control system’s performance. A nonlinear analysis method known as describing functions, based on the assumption that the input to the nonlinearity is sinusoidal, can be used to predict the behavior of a class of nonlinear systems. A nonlinear element does not have a transfer function. However, for a certain class of nonlinearities, it is possible to replace the nonlinearity by a frequency-dependent equivalent gain for analysis purposes. We can then study the properties of the loop, such as its stability. The describing function method is mostly a heuristic method, and its aim is to try to find something akin to a “transfer function” for a nonlinear element. The idea is that in response to a sinusoidal excitation, most nonlinearities will produce a periodic signal (not necessarily sinusoidal) with frequencies being the harmonics of the input frequency. Hence one may

620 Chapter 9 Nonlinear Systems view the describing function as an extension of the frequency response to nonlinearities. We can assume that in many cases we may approximate the output by the first harmonic alone, and the rest can be neglected. This basic assumption means that the plant behaves approximately as a low-pass filter, and this is luckily a good assumption in most practical situations. The other assumptions behind the describing function are that the nonlinearity is time invariant and that there is a single nonlinear element in the system. Indeed, the describing function is a special case of the more sophisticated harmonic balance analysis. Its roots go back to the early studies in the Soviet Union and elsewhere. The method was introduced by Kochenburger in 1950 in the United States. He proposed that the Fourier series be used to define an equivalent gain, Keq (Truxal, 1955, p. 566). This idea has proved to be very useful in practice. The method is heuristic, but there are attempts at establishing a theoretical justification for the technique (Bergen and Franks, 1973; Khalil, 2002; Sastry, 1999). In fact the method works much better than is warranted by the existing theory! Consider the nonlinear element f (u) shown in Fig. 9.24. If the input signal u(t) is sinusoidal of amplitude a, or u(t) = a sin(ωt),

(9.20)

then the output y(t) will be periodic with a fundamental period equal to that of the input and consequently with a Fourier series described by y(t) = a0 +

∞ 

ai cos( jωt) + bi sin( jωt)

(9.21)

i=1

= a0 +

∞ 

Yi sin( jωt + θi ),

i=1

2 π ai = y(t) cos( jωt) d(ωt), (9.22) π 0

π 2 y(t) sin( jωt) d(ωt), (9.23) bi = π 0  Yi = ai2 + bi2 , (9.24)   ai θi = tan−1 . (9.25) bi Kochenburger suggested that the nonlinear element could be described by the first fundamental component of this series as if it were a linear system with a gain of Y1 and phase of θ1 . If the amplitude is varied, the Fourier coefficients and the corresponding phases will vary as a function of the input signal amplitude, due to the nonlinear nature of the element. He called this approximation a describing function (DF). The describing function is defined as the (complex) quantity that is a ratio of the amplitude of the fundamental component of the output of the nonlinear element to the amplitude where

Describing function

Figure 9.24 Nonlinear element

u

f

y

9.4 Equivalent Gain Analysis Using Frequency Response: Describing Functions

621

of the sinusoidal input signal and is essentially an “equivalent frequency response” function: Y1 (a, ω) jθ1 b1 + ja1 Y1 (a, ω) = e = ∠θ1 . DF = Keq (a, ω) = (9.26) a a a Hence the describing function is defined only on the jω axis. In the case of memoryless nonlinearities that are also an odd function [i.e., f (−a) = −f (a)], then the coefficients of the Fourier series cosine terms are all zeros, and the describing function is simply b1 (9.27) DF = Keq (a) = a and is independent of the frequency ω. This is the usual case in control, and saturation, relay, and dead-zone nonlinearities all result in such describing functions. Computation of the describing function for the nonlinear characteristics of Fig. 9.6 is generally straightforward, but tedious. It can be done either analytically or numerically and may also be determined by an experiment. We will now focus on computation of several describing functions for some very common nonlinearities.

EXAMPLE 9.10

Describing Function for a Saturation Nonlinearity A saturation nonlinearity is shown in Fig. 9.25(a) and is the most common nonlinearity in control systems. The saturation function (sat) is defined as ⎧ ⎨ +1, x > 1, |x| ≤ 1, x, sat(x) = ⎩ −1, x < −1. If the slope of the linear region is k and the final saturated values are ±N, then the function is   k y = N sat x . N Find the describing function for this nonlinearity. Solution. Consider the input and output signals of the saturation element shown in Fig. 9.25. For an input sinusoid of u = a sin ωt with amplitude a ≤ Nk , the output is such that the DF is just a gain of unity. With a ≥ Nk , we need to compute the amplitude and phase of the fundamental component of the output. Since saturation is an odd function, all the cosine terms in Eq. (9.21) are zeros and a1 = 0. According to Eq. (9.27), b1 Keq (a) = , a

Figure 9.25 (a) Saturation nonlinearity; (b) input and output signals

u, y

y

u N

N u

h 0 h N

(a)

u

y

0

y t0

t1

N

(b)

t2

t

622 Chapter 9 Nonlinear Systems so that b1 =

2 π





π

N sat 0

 k a sin ωt sin ωt d(ωt), N

since the integral for the coefficient b1 over the interval ωt = [0, π] is simply twice that of the interval ωt = [0, π/2]. Then b1 =

4N π





π 2

sat

0

 k a sin ωt sin ωt d(ωt). N

We can now divide the integral into two parts corresponding to the linear and saturation parts. Let us define the saturation time ts as the time when 1 ts = sin−1 ω



N ak

 or

ωts = sin

−1



N ak

 .

(9.28)

Then 4Nω b1 = πa =

4Nω πa

=

4Nω πa

=

4Nω πa

 

π 2 k a sin ωt sin ωt dt + sat sin ωt dt N 0 ωts  

π ωts k 2 2 a sin ωt dt + sin ωt dt N 0 ωts  

π ωts k 2 a(1 − cos 2ωt) dt + sin ωt dt 2N 0 ωts   k π k 1 ts ts at |0 − a sin 2ωt |0 − (cos − cos ts ) . 2N 2N ω 2 

ωts



But using Eq. (9.28), we have  N sin ωts = ka

and

⎧ ⎨

*N +

cos ωts =



1−

N ka

2 .

We finally obtain Keq (a) =

⎩

 2 π

k sin−1

ak

+ k,

N a



1−

* N +2 ka

 ,

ka N ka N

> 1,

(9.29)

≤ 1.

Figure 9.26 shows a plot of Keq (a) indicating that it is a real function independent of frequency and results in no phase shifts. It is seen that the describing function is initially a constant and then decays essentially as a function of the reciprocal of the input signal amplitude, a.

9.4 Equivalent Gain Analysis Using Frequency Response: Describing Functions Figure 9.26 Describing function for saturation nonlinearity with k = N = 1

623

1.0

Keq

0.8 0.6 0.4 0.2 0

EXAMPLE 9.11

0

1

2

3

4

5 a

6

7

8

9

10

Describing Function for a Relay Nonlinearity Find the describing function for the relay or sgn function shown in Fig. 9.6(b) and defined as sgn(x) = 0, = sign of x,

x = 0, otherwise.

Solution. The output is a square wave of amplitude N for every size input; thus 4N Y1 = 4N π and Keq = π a . The solution can also be obtained from Eq. (9.29) if we let k → ∞. For small angles,   N ∼ N , sin−1 = ak ak and thus, from Eq. (9.27), we have     N N 4N 2 Keq (a) = k + = . (9.30) π ak a πa The preceding two nonlinearities were memoryless. Next we consider a nonlinearity with memory. Nonlinearities with memory occur in many applications, including magnetic recording devices, backlash in mechanical systems, and in electronic circuits. Consider the bistable electronic circuit shown in Fig. 9.27 that is called a Schmitt trigger (Sedra and Smith, 1991). This circuit has memory. Referring to Fig. 9.28, if the circuit is in the state where vout = +N, then positive values of vin do not change the state. To “trigger” the circuit into the state vout = −N, we must 1 make vin negative enough to make v negative. The threshold value is h = NR R2 . The Schmitt trigger is employed commonly in spacecraft control (Bryson, 1994). We next find the describing function for a hysteresis nonlinearity.

EXAMPLE 9.12

Describing Function for a Relay with Hysteresis Nonlinearity Consider the relay function with hysteresis shown in Fig. 9.29(a). Find the describing function for this nonlinearity. Solution. A system with hysteresis tends to stay in its current state. Until the input to the signum function is past the value h, it is not possible to determine the output

624 Chapter 9 Nonlinear Systems Figure 9.27 Schmitt trigger circuit

R1

R2 y

yin

 



yout



Figure 9.28 Hysteresis nonlinearity for Schmitt trigger circuit

Vout N

h

0

Vin

h

N

Figure 9.29 (a) Hysteresis nonlinearity; (b) input and output to the nonlinearity

u, y y

y

N

N

u u

h

0 h N (a)

u

t2

y

t1

t

N (b)

uniquely without knowing its past history. That implies that we have a nonlinearity with memory. The output is a square wave with amplitude N as long as the input amplitude a is greater than the hysteresis level h. From Fig. 9.29(b), it is seen that the square wave lags the input in time. The lag time can be computed as the time when   h a sin ωt = h or ωt = sin−1 . (9.31) a Because the phase angle is known for all frequencies,
   4N 4N −j sin−1 ha −1 h , (9.32) ∠ − sin e = Keq (a) = πa a πa ⎛ ⎞  2 h h⎠ 4N ⎝ . (9.33) 1− −j = πa a a

14 12 10 8 6 4 2 0 0.1

0.2

0.3

0.4

0.5

0.2

0.3

0.4

0.5

a (a)

0.6

0.7

0.8

0.9

1.0

0.6

0.7

0.8

0.9

1.0

625

0 Phase, ⬔Keq, deg

Figure 9.30 Describing function for the hysteresis nonlinearity for h = 0.1 and N = 1: (a) magnitude; (b) phase

Magnitude, Keq

9.4 Equivalent Gain Analysis Using Frequency Response: Describing Functions

20 40 60 80 100 0.1

a (b)

The describing function is then given by ⎧  
2 ⎪ ⎨ 4N h h 1 − a − ja , πa Keq (a) = ⎪ ⎩ 0,

a ≥ h,

(9.34)

a < h.

The describing function is plotted in Fig. 9.30. The magnitude is proportional to the reciprocal of the input signal amplitude, and the phase varies between −90o and 0o .

9.4.1 Stability Analysis Using Describing Functions The Nyquist theorem can be extended to deal with nonlinear systems whose nonlinearities have been approximated by describing functions. In the standard linear system analysis the characteristic equation is 1 + KL = 0, where the loop gain is L = DG and 1 (9.35) L=− . K As described in Section 6.3, we look at the encirclements of the −1/K point to determine stability. With a nonlinearity represented by the describing function, Keq (a), the characteristic equation is of the form 1 + Keq (a)L = 0, and it follows that 1 . (9.36) L=− Keq (a) Now we have to look at the intersection of L with a plot of −1/Keq (a). If the curve L intersects −1/Keq (a), then the system will oscillate at the crossing amplitude, al , and the corresponding frequency, ωl , keeping in mind the approximate nature of the describing function. We then look for encirclements to decide whether the system would be stable for that particular value of the gain, as if it were a linear system. If so, we deduce that the nonlinear system is stable. Otherwise, we infer that the nonlinear system is unstable.

626 Chapter 9 Nonlinear Systems Figure 9.31 Closed-loop system with a nonlinearity

Nonlinear element R






f

y

G(s)



Figure 9.31 shows an example of an otherwise linear system, except for a single nonlinearity. The nonlinear elements may indeed have a beneficial effect and may limit the amplitude of oscillations. The describing function analysis can be used to determine the amplitude and frequency of the limit cycle. Strictly speaking, a limit cycling system can be considered to be unstable. In reality, the trajectory of the limit cycling is confined to a finite region of the state space. If this region is within allowable performance, then the response is tolerable. In some cases, the limit cycling is the beneficial effect (see case study in Section 10.4). The system does not possess asymptotic stability, since the system does not come to a rest at the origin of the state space. The describing function can be beneficial in determining the conditions under which instability results and can even suggest remedies in eliminating instability, as illustrated in the next example, in which the Nyquist plot of the linear loop gain, L, as well as the negative reciprocal of the describing function, −1/Keq (a), are superimposed. The point at which they cross corresponds to the limit cycle. To determine the amplitude and frequency of the limit cycle, we can rewrite Eq. (9.36) as follows: Re{L( jω)}Re{Keq (a)} − Im{L( jω)}Im{Keq (a)} + 1 = 0,

(9.37)

Re{L( jω)}Im{Keq (a)} + Im{L( jω)}Re{Keq (a)} = 0. We can then solve these two equations for the possible two unknown values of the limit-cycle frequency, ωl , and the corresponding amplitude, al , as illustrated in the ensuing examples.

EXAMPLE 9.13

Conditionally Stable System Consider the feedback system in Fig. 9.14. Determine the amplitude and the frequency of the limit cycle using the Nyquist plot. Solution. The Nyquist plot of the system is superimposed on −1/Keq (a) as shown in Fig. 9.32. Note that the negative of the reciprocal of the describing function, using Eq. (9.29), is −

1 = Keq (a)

1

 2 π

k sin

* + −1 N

2 π

+



1



=−

ak

N a

sin

* + −1 0.1 a

+

1−

 0.1 a

* N +2

1−



ka

* 0.1 +2

,

a

which is a straight line that is coincident with the negative real axis and is parameterized as a function of the input signal amplitude, a. The point of the intersection of the two curves at −0.5 corresponds to the limit-cycle frequency of ωl = 1. A plot of the

9.4 Equivalent Gain Analysis Using Frequency Response: Describing Functions Figure 9.32 Nyquist plot and describing function to determine limit cycle

627

Im(G)

v 1 a  0.63

‹– a

1

1  Keq(a)

0.1j

0.5



1 K

0 0.1j

Re(G)

Nyquist plot

1.0 0.8 Keq

Figure 9.33 Describing function for saturation nonlinearity with N = 0.1 and k = 1

0.6 0.4 0.2 0

0

1

2

3

4

5 a

6

7

8

9

10

describing function for k = 1 and N = 0.1 is shown in Fig. 9.33, and a magnitude of Keq = 0.2 corresponds to an input amplitude of al = 0.63. Alternatively, from the root locus of our example shown in Fig. 9.15, the gain at the imaginary-axis crossover is 0.2; then, from Eq. (9.29), we have ⎛ ⎞   2   0.1 2 0.1 ⎠ 0.1 = 0.2. 1− Keq = ⎝sin−1 + π a a a If we approximate the arcsine function by its argument as   0.1 0.1 , sin−1 ≈ a a then ⎛ ⎞     2 2 ⎝ 0.1 0.1 0.1 ⎠ + 1− = 0.2, π a a a which leads to the polynomial equation π 2 a4 − 2πa3 + (0.1)2 = 0 and we find the relevant solution to be a = 0.63. By measurement on the time history of Fig. 9.16, the amplitude of the oscillation is 0.62, which is in good agreement with our prediction.

628 Chapter 9 Nonlinear Systems For systems with nonlinearities that have memory, we can also use the Nyquist technique, as illustrated in the next example.

EXAMPLE 9.14

Determination of Stability with a Hysteresis Nonlinearity Consider the system with a hysteresis nonlinearity shown in Fig. 9.34. Determine whether the system is stable and find the amplitude and the frequency of the limit cycle. Solution. The Nyquist plot for the system is shown in Fig. 9.35. The negative reciprocal of the describing function for the hysteresis nonlinearity is 1 π  2 1  =− = −  a − h2 + jh . −
2 Keq (a) 4N 4N h 1 − − j ha πa a In this case N = 1 and h = 0.1, and we have π  2 1 =− a − 0.01 + j0.1 . − Keq (a) 4 This is a straight line parallel to the real axis that is parameterized as a function of the input signal amplitude a and is also plotted in Fig. 9.35. The intersection of this curve with the Nyquist plot yields the frequency and the corresponding amplitude of the stable limit cycle. We can also determine the limit-cycle information analytically: π  2 1 1 =− . a − 0.01 + j0.1 = G( jω) = − Keq (a) 4 jω( jω + 1) Clearing the denominator in the preceding equation, we have   0.1π 2 π  2 0.1π π 2 ω−1+j ω − a − 0.01ω2 + a − 0.01ω = 0. 4 4 4 4 Setting the real and imaginary parts equal to zero yields two equations and two unknowns. The relevant solution is ωl = 2.2 rad/sec and al = 0.24. A SIMULINK implementation of the closed-loop system is shown in Fig. 9.36. The step response of the system is shown in Fig. 9.37, and the limit cycle has an amplitude of al = 0.24 and a frequency of ωl = 2.2 rad/sec and is well predicted by our analysis.

Figure 9.34 Feedback system with hysteresis nonlinearity

1.0 R






0.1

G(s) 

1 s(s  1)

Y

9.5 Analysis and Design Based on Stability Figure 9.35 Nyquist plot and DF to determine limit-cycle properties

629

Im(G)

0.1714

0 0.0785

‹– a 

Re(G)

v  2.2 a  0.24

1 Keq(a)

a  h  0.1

Nyquist plot

Figure 9.36 SIMULINK
diagram for system with hysteresis

Scope 1 s2  s Transfer fcn

  Step

Figure 9.37 Step response displaying limit-cycle oscillations

Relay

Simout To workspace

1.4 1.2

Output, y

1.0 0.8 0.6 0.4 0.2 0 0.2

0

5

10

15

20

25

30

35

40

45

50

Time (sec)


9.5 Analysis and Design Based on Stability The central requirement of any control system is stability, and the design methods we have studied are based on this fact. The root locus is a plot of closed-loop poles in the s-plane, and the designer is always aware of the fact that if a root strays into the right half of the plane, the system will be unstable. Designs based on the state representation include pole placement, where the desired locations of the poles are, of course, selected to be well within the stable region. In a similar fashion, Nyquist proved conditions for stability based on the frequency response, and designers are aware of the encirclement requirements of their plots or, equivalently, of the gain and phase margins of stability margins in the Bode plots. Prior to either of these methods,

630 Chapter 9 Nonlinear Systems mathematicians studied the stability of ordinary differential equations (ODE), and these and other sophisticated techniques are needed to face the problems of nonlinear systems. We begin with a graphical representation of ODE solutions known as the phase plane and introduce the methods of Lyapunov and others as an introduction to this area of control design.

9.5.1 The Phase Plane Whereas the root locus and the frequency-response methods consider the system response indirectly via either the poles and zeros of the transfer function or the gain and phase of the frequency response, the phase plane considers the time response directly by plotting the trajectory of the state variables. Although direct visualization restricts the method to second-order systems having only two state variables, the ability of the method to consider nonlinearities, as well as to give new insight into linear systems, makes a quick look at the technique well worthwhile. To illustrate the ideas of the phase plane, consider a fictional motor system shown in Fig. 9.38 with the open-loop transfer function 1 . s(Ts + 1) If we assume that T = 1/6 and the amplifier is (for the moment) not subject to saturation and has gain K where K = 5T , the state equations for the closed-loop system can be written as G(s) =

x˙ 1 = x2 ,

(9.38)

x˙ 2 = −5x1 − 6x2 ,

(9.39)

y = x1 .

(9.40)

Because these equations are time invariant, the time can be eliminated by dividing Eq. (9.38) into Eq. (9.39), with the result that dx2 −5x1 − 6x2 = . (9.41) dx1 x2 The solution to this equation gives a plot of x2 versus x1 or, in other words, a trajectory in the phase plane of coordinates (x1 , x2 ).6 Before plotting Eq. (9.41), it is useful to consider the system equations first in matrix form as x˙ = Fx for which   0 1 F= . −5 −6 Figure 9.38 An elementary position feedback system with a nonlinear actuator

r








e

u 




1 Ts

x2

1 s

x1

y

6 If the slope dx /dx is set to a constant, the relation between x and x is a straight line. If the known 2 1 2 1 values of slopes are marked along a set of these lines, the trajectories can be readily sketched. For example, along the x1 axis, where x2 = 0 the slope is ∞ and the trajectories are vertical. This method is called the method of isoclines.

9.5 Analysis and Design Based on Stability

631

If in this equation we assume that x = xo est , where both s and xo are constants, then x˙ = xo sest , and the equation can be reduced as follows: x˙ = Fx,

(9.42)

xo sest = Fxo est ,

(9.43)

[sI − F]xo e = 0,

(9.44)

[sI − F]xo = 0.

(9.45)

st

Here it should be recognized that Eq. (9.45) is the eigenvector equation for the matrix F, which, in component form, is      0 s −1 x01 = . (9.46) x02 0 5 s+6 As described in Appendix WE, Eq. (9.46) has a solution only if the determinant of the coefficient matrix is zero, for which s(s + 6) + 5 = 0,

(9.47)

(s + 1)(s + 5) = 0.

(9.48)

The two values of s for which the equation has a solution are the eigenvalues s = −1 and s = −5. If we substitute s = −1 into Eq. (9.46), we obtain      −1 −1 x01 0 = , (9.49) 5 −1 + 6 x02 0 from which the solution for the initial state vector is x02 = −x01 . This line in the state space is the eigenvector corresponding to the eigenvalue s = −1. If we repeat this process with s = −5, the result is      −5 −1 x01 0 = , (9.50) x02 5 −5 + 6 0 and in this case the solution for the eigenvector is x02 = −5x01 . Consider what all this means. We started with the assumption that the time solution for the state is a constant times an exponential. We found that this is possible only if the exponential is either e−t or e−5t . In the first case, the state must lie along the vector x02 = −x01 , and in the second, the state must lie along the vector x02 = −5x01 . With this knowledge, we compute the solutions to Eq. (9.38) and Eq. (9.39) for different initial conditions and plot x1 (t) vs. x2 (t) in Fig. 9.39. In the figure, the two eigenvectors are identified. When we look at these curves, it is clear that all the paths start parallel to the (fast!) eigenvector corresponding to s = −5 and quickly move to the (slow!) one corresponding to s = −1. All trajectories approach the equilibrium point at the origin of the state space. The plot will be substantially changed if the amplifier saturates. For example, if the amplifier saturates at a value of u = 0.5, then the velocity, x2 , will rapidly approach this value and will be stuck there until the position reaches a value that brings the amplifier out of saturation. The new plot is shown in Fig. 9.40. Notice that in the linear region the motion is almost entirely along the slow eigenvector. Finally we note that the phase-plane portrait changes again when the poles are complex. In that case, the motion of the state variable is composed of damped

632 Chapter 9 Nonlinear Systems Figure 9.39 Phase-plane plot of a node with poles at s = −1 and s = −5

1.0 0.8

s  5

s  1

x02  x01

0.6 0.4

x2

0.2 0 0.2

x02  x01

0.4 0.6 0.8 1.0 1.5

Figure 9.40 Phase-plane plot with saturation

1.0

0.5

0 x1

0.5

1.0

1.5

2.0 1.5 1.0

x2

0.5 0 0.5 1.0 1.5 2.0 8

6

4

2

0 x1

2

4

6

8

sinusoids and the plot of x1 versus x2 is along a spiral. A collection of trajectories for various initial conditions is shown in Fig 9.41. These few examples just scratch the surface of phase-plane analysis but give some idea of the use of this format in helping a designer to visualize dynamic responses.

Bang-Bang Control One example of design for a nonlinear system based on the phase plane is that of optimal minimum time control in the face of control saturation. For our purposes here, the simplest version of this widely used technique is introduced: that of the 1/s2

9.5 Analysis and Design Based on Stability Figure 9.41 Phase-plane plot for a system with complex poles

633

Phase plane for a focus 4 3 2

x2

1 0 1 2 3 4 1.5

1.0

0.5

0 x1

0.5

1.0

1.5

plant. The equations are y¨ = u,

(9.51)

e = y − yf ,

(9.52)

where y(0) = y˙ (0) = 0, yf is a constant, and the control is constrained to be |u| ≤ 1. The problem is to drive the error to be identically zero in minimum time. If we define state variables as x1 = e and x2 = e˙ = y˙ , the equations reduce to x˙ 1 = x2 ,

(9.53)

x˙ 2 = u,

(9.54)

x1 (0) = −yf ,

(9.55)

x1 (tf ) = x2 (tf ) = 0,

(9.56)

and the problem is to minimize tf . Intuitively, this is the problem of the eager driver who wishes to speed from one stop to the next in minimum time. She would put the pedal to the metal for a time and then switch to stand on the brakes as the car skids to a stop at just the right place. A basic result of the theory of optimal control confirms this intuitive idea that the solution to this problem is, if yf > 0, to apply full positive control for a time and then to switch to full negative control at just the right time to cause the error to reach the origin and stop there. To study the case, a plot of the trajectories of the plant in the phase plane for the two cases of u = 1 and u = −1 is given in Fig. 9.43. For u = +1, the trajectories start in the fourth quadrant and move up to the first. For u = −1, they start in the second quadrant and move down to the third. Two segments of this family are of particular interest: those that pass through the origin. Once the path reaches one of these, a constant control will bring the state to the desired final resting place. Therefore, for any initial condition, once the trajectory reaches one of the two curves going through the origin, the correct action is to switch

634 Chapter 9 Nonlinear Systems Figure 9.42 Phase-plane of the 1/s2 plant for ±1 controls

25 20

u  1

15 10

x2

5 0 5 10 15

u  1

20 25 150

Figure 9.43 Switching curve for the 1/s2 plant

100

50

0 x1

50

100

150

15

10

x2

5

0 5 10 15 100 80 60 40 20

0 x1

20

40

60

80

100

the control (u = +1 to −1, or u = −1 to +1) so that the trajectory will follow that curve to the origin. The “switching curve” is plotted in Fig. 9.44. For a second-order plant, the switching curve can be found by reversing the time in the equations of motion, setting the initial state to zero and applying the maximum control. The process can be repeated with minimum control to sweep out the other branch of the curve. For any initial condition above the curve, u = −1 is applied and for any initial condition below the curve, u = +1 is used. As described, the result will be a minimumtime response. Notice that the curve has vertical slope at the origin; as a result, the implementation is extremely sensitive in this neighborhood. A modified version known as the proximate time-optimal system (PTOS) used in the computer disk drive industry was studied by Workman (1987). The modification consists of shifting the

9.5 Analysis and Design Based on Stability Figure 9.44 Response of a time-optimal system

635

Control for the time-optimal system 1.0 u

0.5 0 0.5 1.0 0

1

2

3

4

5

6

Time (sec) Output for the time-optimal system 5

y

4 3 2 1 0

Figure 9.45 Response of a PTOS system

0

1

2 3 Time (sec)

4

5

4

5

4

5

Control for the PTOS, a  0.9 1.0 u

0.5 0 0.5 1.0 0

1

2 3 Time (sec) Output for the PTOS

y

4 3 2 1 0

0

1

2 3 Time (sec)

curves a bit and replacing the infinite slope at the origin with a finite-slope, linear control region. The result has been widely used for hard-disk drives and similar systems. Typical responses for a time-optimal system and for a PTOS system generated with SIMULINK are given in Figs. 9.45 and 9.46. Notice that the response times are almost exactly the same, but while the time optimal system control has a violent

636 Chapter 9 Nonlinear Systems Figure 9.46 SIMULINK
diagram for position feedback system

r Step

 

e

u

Saturation

 

x2

1 s Integrator1

1 s

x1 y Scope

Integrator

chatter at the end where the switching curve has infinite slope, the output of the PTOS system slips into its final value smoothly. For a more exact study, we need to turn to the nonlinear equations.

9.5.2 Lyapunov Stability Analysis

Stability in the sense of Lyapunov

The stability of motion as studied by Lyapunov involves sophisticated mathematics beyond the scope of this text. Here we will present heuristic arguments giving the flavor of the theory and state a few of the most basic results. Lyapunov presented two methods for the study of stability of motion described by systems of ODEs. His indirect or first method is based on linearization of the equations and drawing conclusions about the stability of the nonlinear system by considering the stability of the linear approximation. He proved the results of the first method by use of his direct or second method, in which the nonlinear equations are considered directly. A discussion of the indirect method serves to introduce both methods. The problem requires a new definition of stability suitable for the vector-matrix equations. Intuitively, we say a system is stable if initial conditions of moderate size result in a response that remains of moderate size. To express this mathematically, first we need a definition of “size.” This is the norm of a vector for which the symbol is x. Of the many possible definitions, we  select here the familiar Euclidean measure defined by its square as x2 = xT x = ni=1 xi2 . With this idea, the definition of stability used is that if one is given a sphere of any radius , one can find a smaller sphere of radius δ such that if the initial state is inside δ, then the trajectory will, for all time, remain inside . A bit more formally, the system is stable if, for any given  > 0, one can find a δ > 0 such that if ||x(0)|| < δ, then ||x(t)|| <  for all t. If the state is not only stable, but in the limit as t → ∞, ||x(t)|| → 0, the system is said to be asymptotically stable. If, for any , it is possible to select δ arbitrarily large, then the system is said to be stable in the large. Study of these matters begins with the time-invariant ODE equation x˙ = f(x),

(9.57)

x˙ = Fx + g(x).

(9.58)

for which the linear approximation is

In this equation, it is assumed that all the linear terms are in Fx and higher-order terms are in g(x), in the sense that when x gets small, g(x) gets small faster, as expressed by lim

x→0

g(x) = 0. x

(9.59)

Lyapunov’s second method begins with the intuitive notion that one measure of the size of the state of a physical system is the total energy stored in the system at any instant

9.5 Analysis and Design Based on Stability

Lyapunov function

637

and the observation that when the stored energy is no longer changing, the system must be at rest. For an electric circuit, for example, the electric energy is proportional to the square of the capacitor voltages and the magnetic energy is proportional to the square of the inductor currents. Lyapunov extracted the abstract essence of this idea and defined a scalar function of the state V (x), called a Lyapunov function, having the following properties: 1. V (0) = 0; 2. V (x) > 0, x = 0; 3. V is continuous and has continuous derivatives with respect to all components of x; ∂V 4. V˙ (x) = ∂V ∂x x˙ = ∂x f(x) ≤ 0 along trajectories of the equation. The first three conditions ensure that in a neighborhood of the origin the function is like a smooth bowl sitting at the origin of the state space. The fourth condition, which obviously depends on the equations of motion, ensures that if δ is selected so that the initial conditions are deeper in the bowl than any part of the ball defined by , the trajectory never climbs higher on the bowl than it was at the start and so remains within , and the system will be stable. Furthermore, if condition 4 is strengthened to be V (x) < 0, then the value of the function must fall to zero and, by condition 1, the state also goes to zero. The stability theorem, which is the basis for Lyapunov’s second method, states that

Lyapunov’s second method

If a Lyapunov function can be found for a system, then the motion is stable and, furthermore, if V (x) < 0, the motion is asymptotically stable. The second method is to search for a Lyapunov function. The hard part for the application of this theory is the statement “If a Lyapunov function can be found.” Only in the linear case is a prescription given for finding a Lyapunov function; otherwise the theory only gives the engineer a hunting license to look for such a function. We are now in a position to consider the indirect method for stability of Eq. (9.58). Perhaps because energy in simple systems is a sum of the squares of the variables, for this problem Lyapunov considers a quadratic candidate for V by supposing that a symmetric positive definite matrix P can be found and the function defined as V (x) = xT Px. Clearly the first three conditions are satisfied by this function; the fourth condition must be tested before it can be concluded that we have a Lyapunov function. The calculation of V˙ is V (x) = xT Px,

(9.60)

V˙ (x) = x˙ T Px + xT Px˙

(9.61)

= (Fx + g(x)) Px + x P(Fx + g(x))

(9.62)

= x (F P + PF)x + 2x Pg(x).

(9.63)

T

T

T

T

T

A basic matrix result, known as a Lyapunov equation, is FT P + PF = −Q,

(9.64)

638 Chapter 9 Nonlinear Systems and he showed that if F is a stability matrix having all its eigenvalues in the LHP, then for any positive definite matrix Q, the solution P of this equation will also be positive definite. The argument from here is to select Q and solve for P. Then, if the eigenvalues of F are in the LHP, P will be positive definite, so V (x) is a possible Lyapunov function and (9.65) V˙ (x) = −xT Qx + 2xT Pg.

Lyapunov’s first method

The final part of the argument is to note that, by Eq. (9.59), if x is small enough, then the first term of Eq. (9.65) will dominate, the fourth condition is satisfied, V is a Lyapunov function, and the system has been proven to be stable. Note that the requirement that x be small enough guarantees only that there is a neighborhood of the origin which is stable. Further conditions are needed to show that the bowl defined by V extends to ∞ in all directions as ||x|| tends to ∞ (and not before!), so that stability holds for all states and is “in the large.” There is also an instability theorem which shows that if any eigenvalue of F is in the RHP, then the origin will be unstable. If all the poles of F are in the LHP except for some simple poles on the imaginary axis, then stability depends on further properties of the nonlinear terms, g(x). With this result in hand, the first or indirect method of Lyapunov can be stated: 1. Find the linear approximation and compute the eigenvalues of F. 2. If all the eigenvalues are in the LHP, then there is a region of stability about the origin. 3. If at least one of the eigenvalues is in the RHP, then the origin is unstable. 4. If there are simple eigenvalues on the imaginary axis and all other values are in the LHP, then no statement about stability can be made based on this method.

EXAMPLE 9.15

Lyapunov Stability for a Second-Order System Use Lyapunov’s method to find conditions for the stability of a second-order linear system described by the state matrix   −α β F= . −β −α Solution. For the linear case we can take any positive definite Q we like; the simplest is Q = I. The corresponding Lyapunov equation is         −α −β p q p q −α β −1 0 + = . (9.66) β −α q r q r −β −α 0 −1 Multiplying out Eq. (9.66) and equating coefficients, we get −αp − βq − αp − βq = −1,

(9.67)

−αq − βr + βp − αq = 0,

(9.68)

βq − αr + βq − αr = −1.

(9.69)

9.5 Analysis and Design Based on Stability

639

Equations (9.67) to (9.69) are readily solved to get p = r = 1/2α, q = 0, so that   1 0 2α P= , 1 0 2α and the determinants are 1/2α>0 and 1/4α 2 >0. Thus P > 0, so we conclude that the system is stable if α > 0. For systems with many state variables and nonnumeric parameters, solution of the Lyapunov equation can be burdensome, but the result is an equivalent alternative to Routh’s method for computing the conditions for stability in a system with literal parameters.

EXAMPLE 9.16

Lyapunov’s Direct Method for a Position Feedback System Consider the position feedback system modeled in Fig. 9.38. Illustrate the use of the direct method on this nonlinear system. Simulate the system using SIMULINK, assuming that T = 1, and evaluate the step response of the system. Solution. We assume that the actuator, which is perhaps only an amplifier in this case, has a significant nonlinearity, which is shown in the figure as a saturation but is possibly more complex. We will assume  e only that u = f (e), where the function lies in the first and third quadrants so that 0 f (e) de > 0. We also assume that f (e) = 0 implies that e = 0, and we will assume that T > 0, so the system is open-loop stable. The equations of motion are e˙ = −x2 , 1 f (e) . x˙ 2 = − x2 + T T

(9.70a) (9.70b)

For a Lyapunov function, consider something like kinetic plus potential energy:

e T V = x22 + f (σ ) dσ . (9.71) 2 0 Clearly, V = 0 if x2 = e = 0 and, because of the assumptions about f , V > 0 if x22 + e2 = 0. To see whether the V in Eq. (9.71) is a Lyapunov function, we compute V˙ as follows: V˙ = Tx2 x˙ 2 + f (e)˙e   1 f (e) = Tx2 − x2 + + f (e)(−x2 ) T T = −x22 . Hence V˙ ≤ 0 and the origin is Lyapunov stable. Moreover, V˙ is always decreasing if x2 = 0, and Eq. (9.70b) indicates that the system has no trajectory with x2 ≡ 0, except x2 = 0. Thus we can conclude that  the system is asymptotically stable for every f that satisfies two conditions: (1) f dσ > 0 and (2) f (e) = 0 implies that

640 Chapter 9 Nonlinear Systems Figure 9.47 Step response for position control system

1.2 1

Output

Amplitude

0.8 0.6 0.4 0.2 0 0.2

x2 0

5

10

15

Time (sec)

e = 0. The SIMULINK diagram for the system is shown in Fig. 9.46 for T = 1. The step response of the system is shown in Fig. 9.47. As we mentioned earlier, the study of the stability of nonlinear systems is vast, so we have only touched here on some important points and methods. Further material for study can be found in LaSalle and Lefschetz (1961), Kalman and Bertram (1960), Vidyasagar (1993), Khalil (2002), and Sastry (1999).

Lyapunov Redesign of Adaptive Control One of the classical applications of Lyapunov stability theory to control is a technique known as Lyapunov redesign. The idea is to construct the system with some key control parameters unspecified, propose a candidate Lyapunov function and then select the available components to force the candidate to succeed and be an actual Lyapunov function from which stability can be concluded. The method was applied in an early paper by Parks (1966) to a model reference adaptive control system. A block diagram of the simple system first considered is drawn in Fig. 9.48. In this system, the model and the plant have the same dynamics but different gains. The objective is to adjust the control gain, Kc , so that Kc Kp = Km , and the plant output, yp , will equal the model output, ym . A proposed heuristic rule, known as the “MIT” rule, was based on the idea that if we define the cost as the square of the instantaneous error and move Kc so as to make this cost smaller, the result should drive Kc to the right value. If the gradient of the cost is positive (pointing uphill, so to speak), the gain should be reduced, and if the gradient is negative, the gain should be increased. Thus the time derivative of the gain should be proportional to the negative Figure 9.48 Block diagram of a simple model reference adaptive system

Km

1 Ts1

ym



R KcKp

1 Ts1

 yp




E

9.5 Analysis and Design Based on Stability

641

of the gradient. In equation form, J = e2 , (9.72) ∂J ∂e = 2e , (9.73) ∂Kc ∂Kc ∂e dKc = −Be , (9.74) dt ∂Kc where B is the “adaptive gain” to be chosen. From the block diagram, Kc K p − K m R(s), (9.75) E(s) = Ts + 1 Kp ∂E R (9.76) = ∂Kc Ts + 1 Kp = Ym . (9.77) Km If we substitute the result of Eq. (9.77) into Eq. (9.74), the result is the MIT rule, dKc = −B
eYm , (9.78) dt where there is a new adaptive gain, B
. Unfortunately, the stability of this rule is not established, and some analysis showed that it could be unstable under reasonable circumstances, such as if there are unmodeled dynamics or disturbances. Parks proposed that Lyapunov redesign would be a better idea and also proposed that, rather than taking K˙ c given by Eq. (9.74), this choice be made in a way that guarantees stability. His idea begins with the differential equations where r = ro a constant: T e˙ + e = (Kc Kp − Km )ro , K˙ c = −B eYm .


(9.79) (9.80)

To simplify things, the definition is made x = (Kc Kp − Km ) and x˙ is to be found. Parks selects V = e2 + λx 2 as a candidate Lyapunov function and computes V˙ = 2e˙e + 2λx x˙ (9.81)
xr  e o = 2e − + 2λx x˙ . (9.82) T T e2 o ˙ If x˙ in the last equation is selected to be x˙ = − er λT , then V = −2 T , the conditions for a Lyapunov function are satisfied, and stability is assured for the given assumptions. Working back, we find that the new algorithm is K˙ c = −B

ero . (9.83) Obviously, this result does not answer the questions of unmodeled dynamics or disturbances, but the principle is clear: Leaving key control equations to be defined so as to obtain a Lyapunov function can put the stability of a system on a firm foundation. As a second example of Lyapunov redesign, consider the adaptive control of a motor shown in Fig. 9.49. Defining the model output as ym and the plant output as yp , the equations are y¨ m + 2ζ ωn y˙ m + ωn2 ym = ωn2 r, y¨ p + 2ζ ωn y˙ p + ωn2 yp

=

Kc Kp ωn2 (r

(9.84) − yp ) + ωn2 yp .

(9.85)

642 Chapter 9 Nonlinear Systems Figure 9.49 Block diagram of adaptive control of a motor

wn^2(s) s2  2*z*wn^2s  wn^2 Model


Error

Input

Kc(s)

Kp*wn^2(s)

1

s2  2*z*wn^2s

Controller

Plant




(In the equation for yp the term ωn2 yp as been added to both sides to make the error equation simpler.) The error is defined as e = ym − yp and an equation for error can be obtained by subtracting the equation for yp from that for ym . The result is e¨ + 2ζ ωn e˙ + ωn2 ye = ωn2 (1 − Kc Kp )(r − yp ).

(9.86)

The idea now is to find an equation for Kc that will result in a Lyapunov function for the error equation. To simplify the calculation, we define the parameter as x = 1 − Kc Kp , for which x˙ = −Kp K˙ c and in terms of which the error equation is e¨ p + 2ζ ωn e˙ + ωn2 e = ωn2 x(r − yp ).

(9.87)

At this point, Parks suggests consideration of V = e2 + α˙e2 + βx 2 as a possible function. We need to find x˙ so this V will be a Lyapunov function. The equation for the derivative is V˙ = 2e˙e + 2α˙ee¨ + 2βx x˙ (9.88) = 2e˙e + 2α˙e{−2ζ ωn e˙ − ωn2 e + ωn2 x(r − yp )} + 2βx x˙

(9.89)

= −4αζ ωn e˙ 2 + 2e˙e(1 − αωn2 ) + x{2α˙eωn2 (r − yp ) + 2β x˙ }.

(9.90) If we take 1 − = 0 and − yp ) + 2β x˙ = 0, then the equation for V˙ 2 ˙ simplifies to V = −4αζ ωn e˙ , which is always negative, and V is a Lyapunov function and the system is stable. Substituting for x, we get the adaptive control law K˙ c = −β
e˙ (r − yp ), (9.91) αωn2

2α˙eωn2 (r

αω2

where β
is a new constant equal to Kp βn . Clearly, we have only touched on Lyapunov’s theory of stability, and our examples of redesign are ancient history from 1966, but they illustrate the principle very well and give a good start to further study of this important area.

9.5.3 The Circle Criterion A nonlinear system with only one single-input-single-output nonlinearity may be represented as shown in Fig. 9.50 by drawing the block diagram from the points of the input and output of the nonlinearity. In the literature this is referred to as the Lur’e problem after the Soviet scientist who first studied it. We assume that the system is unforced and thus r ≡ 0. It is possible to derive a graphical sufficient condition for stability of such systems. Even though this method is practical, it may lead to conservative results in some cases, although extensions exist that yield less conservative results (see Safonov, et al., 1987). First we define sector conditions for memoryless nonlinearities.

9.5 Analysis and Design Based on Stability Figure 9.50 Block diagram of a nonlinear system

r
0






y

G(s)

 Time-varying nonlinearity

f(t, y)y

Figure 9.51 Output of the nonlinearity confined in a sector

643

y

f(t, y)y Slope k2

Slope k1

y

Sector Conditions A function f (x) with a scalar input and a scalar output belongs to the sector [k1 , k2 ] if, for all inputs x, (9.92) k1 x 2 ≤ f (x)x ≤ k2 x 2 . This relationship may be rewritten as f (x) ≤ k2 x = 0. (9.93) x Basically, the definition says that the graph of f (x) lies between two straight lines of slopes k1 and k2 going through the origin, as shown in Fig. 9.51. In this definition, k1 and k2 are allowed to be −∞ or +∞. Note that the sector conditions place no limits on the incremental gain or slope of the function f (x). The ensuing examples illustrate how k1 and k2 are determined. k1 ≤

EXAMPLE 9.17

Computation of a Sector for Signum Nonlinearity Determine a sector that contains the signum function y = f (u) shown in Fig. 9.6(b). Solution. Since sgn(0) = 0, we know that the only line going through the origin that bounds the signum function from above is the y-axis, corresponding to a slope of k2 = ∞. Similarly the line going through the origin that bounds the signum function from below has a slope of zero and corresponds to the x-axis; therefore, k1 = 0. Hence the sector for the signum function is [0, ∞].

644 Chapter 9 Nonlinear Systems Figure 9.52 Sector for saturation

Output

0.1

0.1

EXAMPLE 9.18

Input

Sector for a Saturation Nonlinearity Consider the saturation nonlinearity shown Fig. 9.52. Determine a sector for this function. Solution. The function is bounded above by a line of slope 1, k2 = 1, and is bounded below by the x-axis, k1 = 0, as shown in the figure. Therefore, the sector for this function is [0, 1].

Circle Criterion In 1949 the Russian scientist Aizermann conjectured that if a Lur’e system is stable with f replaced by any linear gain between the limits k1 < k < k2 , then the system will be stable, with the gain replaced by a nonlinearity in the sector [k1 , k2 ]. That means that if a single-loop (strictly proper) continuous-time feedback system as shown in Fig. 9.50 with a linear forward path (F, G, H) is stable for all linear fixed feedback gains k in the range k1 < k < k2 , such that the resultant closed-loop system matrix F + kGH is stable, then the nonlinear system having a memoryless nonlinear timevarying feedback term f (t, y) in the sector [k1 , k2 ], shown in Fig. 9.50, is also stable. Unfortunately, this conjecture is not true as counterexamples exist.7 However, a variation of Aizermann’s conjecture is true and is known as the circle criterion. Rather than give a rigorous proof of the criterion, we describe a heuristic argument that gives insight into the problem and motivates the proof. An electric circuit with a linear impedance, Z( jω) = R(ω) + jX(ω), is described by Ohm’s law as V = IZ(s). We assume that Z is composed of real components, which means that the real part R 7Aizermann’s conjecture spurred a lot of research in this area and led to the development of the Kalman–

Yakubovich–Popov lemma, giving state-space conditions for a passive system. The lemma is used in a proof of the circle criterion.

9.5 Analysis and Design Based on Stability

645

is even and the imaginary part X is odd; that is R(−ω) = R(ω) and X(−ω) = −X(ω). If R(ω)  δ > 0 for all ω, the impedance is called strictly passive. It will dissipate energy. The instantaneous power into the circuit is p = v(t)i(t), and the total energy ∞ absorbed by the circuit is e = 0 v(t)i(t) dt. Referring to the figure, Ohm’s law is equivalent to the plant equation Y = UG(s) with Y as voltage, U as current, and G(s) = R + jX as the impedance. Applying the expression for energy to the plant equation and using the theorem by Parseval8 to convert this to the frequency domain yields

∞

∞ 1 y(t)u(t) dt = U( jω)Y (−jω) dω (9.94) 2π −∞ 0

∞ 1 U( jω)U(−jω)G(−jω) dω (9.95) = 2π −∞

∞ 1 |U( jω)|2 (R − jX) dω (9.96) = 2π −∞

∞ 1 |U( jω)|2 R(ω) dω. (9.97) = 2π −∞ In the last step, the fact the X is odd was used. At this point, the use of more or less conventional notation will simplify the equations substantially. We define inner products and norms as

∞ y(t)u(t) dt = < y, u >, (9.98) 0

u2 =

∞

[u(t)]2 dt =< u, u > .

(9.99)

0

With this notation, and with the assumption that R  δ > 0, Eq. (9.97) is reduced to < y, u >  δ u2 .

(9.100)

Turning now to the nonlinear component, using the same concept of “energy” and assuming that f is in the sector [0, K], we have

∞

∞ [f (y, t)]2 y(t)f (y, t) dt = dt (9.101) f (y, t) 0 0 y(t) 

f (y, t)2 K

(9.102)



u(t)2 . K

(9.103)

The assumption now is that if the total energy given by the sum of Eq. (9.100) and Eq. (9.102) is positive, then the system must be stable, as energy is being steadily lost. The actual value of the energy lost would be equal to the initial energy stored in the 8 See Appendix A.

646 Chapter 9 Nonlinear Systems elements of the system. From this one would conclude that if δu2 + then the system is stable. Thus the criterion is u(t)2 δ u2 + > 0, K   1 u(t)2 > 0, Re{G( jω)} + K Re{KG( jω) + 1} > 0.

u(t)2 > 0, K (9.104) (9.105) (9.106)

In deriving Eq. (9.106), the assumption was made that the nonlinearity was in a zero sector, [0, K]. If the function is actually in the sector [k1 , k2 ], it can be reduced to a zero sector by adding and subtracting k1 in the block diagram as shown in Fig. 9.53. With this change, the dynamic system is replaced by H = 1+kG1 G and the function by f
= f − k1 , which is in the sector [k2 − k1 , 0]. With these changes, the stability criterion is transformed to  G Re 1 + (k2 − k1 ) > 0, 1 + k1 G   1 + k1 G + (k2 − k1 )G > 0, Re 1 + k1 G   1 + k2 G( jω) Re > 0. 1 + k1 G( jω) 

(9.107) (9.108) (9.109)

2 G( jω) It is a fact that a bilinear function such as F = 1+k 1+k1 G( jω) in Eq. (9.109) will map a circle in the F plane into another circle in the G plane (see Appendix WD). In this case, the acceptable region is Re{F} > 0, of which the boundary is the imaginary axis, so the map is from the imaginary axis, a circle of infinite radius, into a finite circle. Because the functions are real, the circle must be centered on the real axis and we need only locate the two points on the real axis. For example, when F = 0, we have 1 + k2 G = 0 or G = − k12 . The other point on the real axis is when the function is infinite, at which point 1 + k1 G = 0 or G = − k11 . Thus the circle in the G plane  is centered on the real axis and goes through the points − k12 , − k11 as plotted in

Figure 9.53 Block diagram manipulation for sector

 

u(t)

y(t)

G(s) 1

 

Repeating sequence stair

Transfer fcn k1 Gain

v(t)  f(y)  

k1 Gain1 f(u) Function

9.5 Analysis and Design Based on Stability Figure 9.54 Illustration of circle criterion

647

Im(L)



1 k1



1 k2

Re(L)

Fig. 9.54. Since F had to avoid the LHP, if we set F = −1, which is in the forbidden region, and solve, we find that G = k1−2 +k2 , which is inside the circle, from which we conclude that the system will be stable if the plot of G( jω) avoids this circle. The actual theorem is as follows: The nonlinear system described is asymptotically stable given that 1. f (t, y) lies in the sector [k1 , k2 ] with 0 ≤ k1 < k2 and 2. the Nyquist plot of the transfer function G( jω) = H( jωI − F)−1 G does not intersect or encircle the “critical circle,” which is centered on the real axis and passes through the two points −1/k1 and −1/k2 , as shown in Fig. 9.54. Circle criterion

In effect, the usual Nyquist “−1” point is replaced by the critical disk. This result is known as the circle criterion or circle theorem and is due to Sandberg (1964) and Zames (1966). Note that these conditions are sufficient but not necessary, because intersection of the transfer function G(s) with the circle as defined does not prove instability. The critical circle is centered at   1 k1 + k2 1 1 − − , =− c= 2 k1 k2 2k1 k2 and has a radius of

k2 − k1 . 2k1 k2 If k1 = 0, then the critical circle degenerates into a half plane defined by Re {G}  −1/k2 . The circle criterion and the describing function are related. In fact, for the case of time-invariant odd nonlinearities that are within a sector and whose describing functions are real, the describing function satisfies the relationship k1 ≤ Keq (a) ≤ k2 so that

for all a,

(9.110)

1 1 1 ≤− , ≤− (9.111) k1 Keq (a) k2 and the plot of the negative reciprocal of the describing function will lie inside the critical circle. This can be seen by the following lower and upper bounds:

π

2 2k1 π 2 Keq (a) = f (a sin(ωt)) sin(ωt) d(ωt) ≥ sin (ωt) d(ωt) = k1 , πa 0 π 0 (9.112) −

648 Chapter 9 Nonlinear Systems Figure 9.55 Nyquist plot and circle criterion

Im(G)

v 1 a  0.63 1

0.1j

0.5

0 0.1j

Re(G)

Nyquist plot

Keq (a) =

2 πa



π

f (a sin(ωt)) sin(ωt) d(ωt) ≤

0

2k2 π



π

sin2 (ωt) d(ωt) = k2 .

0

(9.113) The equivalent gain analysis and describing functions yield the same results. If we take the gain of the describing function, then the amplitude of the limit cycle can be predicted as with the describing functions. Both equivalent gain techniques can be used to determine stability, but as we have seen, the circle criterion allows for time-varying nonlinearities.

EXAMPLE 9.19

Determination of Stability Using the Circle Criterion Consider the system in Example 9.7. Determine the stability properties of the system using the circle criterion. Solution. The related sector is the same one found in Example 9.18. The critical circle degenerates into a half plane defined by Re(G) ≤ −1, as shown in Fig. 9.55. Since the Nyquist plot lies entirely to the right of the critical circle, the system is stable.

9.6 Historical Perspective Almost all physical dynamic systems are nonlinear; hence it is not surprising that the study of nonlinear systems has a long and rich history. The study of nonlinear systems goes back to astronomy and the study of the stability of the solar system dating back to Torricelli (1608–1647), Laplace, and Lagrange. The field got a jolt of “energy” with the doctoral dissertation of A. M. Lyapunov in Russia in 1892. He was trying to solve the stability of rotating bodies of fluids posed by Poincaré and recognized that if he could show that system stored energy was always decreasing, then the system would be stable and eventually come to rest. The study of Lyapunov

Summary

649

functions was introduced to the control field in 1960 by Kalman and Bertram and has evolved rapidly since then. Maxwell was the first to study stability by linearization about an equilibrium point by the derivation of the linear model for the Watt’s fly-ball governor and stating that the system will be stable if the characteristic roots have negative real parts. Kochenberger derived the describing function method in an attempt to handle nonlinearities in 1950 based on frequency-response ideas. Lur’e proposed the absolute stability problem in 1944 and in 1961 Popov developed the circle criterion for nonlinear stability analysis. Yakubovich (1962) and Kalman (1963) later established connections between Lur’e and Popov’s results. The study of adaptive control received a lot of attention during the three decades of the 1960s, 1970s, and 1980s. Adaptive controllers are both time varying and nonlinear in general. During the 1960s sensitivity methods and the MIT rule for adaptive adjustments were developed by Draper and others. Methods to study adaptive systems based on Lyapunov’s methods and passitivity were developed in the 1970s. Robust adaptive control methods were studied in the 1980s. Also, there has been a lot of research on systems, such as the weather, where a minute change in initial conditions or parameters can cause drastic changes in the response of the system. Such systems are said to be chaotic. In all recent studies of nonlinear systems, the availability of powerful computers to solve the equations and to graph the results has been critical. Development of a general theory of nonlinear control continues to be a dream of control theorists and is an ongoing quest.

SUMMARY • • • • • • • • • •

The nonlinear equations of motion may be approximated by linear ones by considering a small-signal linear model that is accurate near an equilibrium. In many cases, the inverse of a nonlinearity may be used to linearize a system. Nonlinearities with no dynamics, such as saturation, can be analyzed using the root locus by considering the nonlinearity to be a variable gain. The root-locus technique can be used to determine the limit-cycle properties for memoryless nonlinearities, and yields the same results as the describing function. The describing function is essentially a heuristic method with the goal of finding a frequency-response function for a nonlinear element. The stability of systems with a single nonlinearity can be studied using the describing function method. The describing function can be used to predict periodic solutions in feedback systems. The Nyquist plot together with the describing function can be used to determine limit-cycle properties. The stability of a nonlinear system in state-space description can be studied by the methods of Lyapunov. The circle criterion provides a sufficient condition for stability.

650 Chapter 9 Nonlinear Systems

REVIEW QUESTIONS 1. Why do we approximate a physical model of the plant (which is always nonlinear) with a linear model? 2.

How would you linearize the nonlinear system equation for radiation heat transfer T˙ = T 4 + T + u?

3. A lamp used as a thermal actuator has a nonlinearity such that the experimentally measured output power is related to the input voltage by P = V 1.6 . How would you deal with such a nonlinearity in feedback control design? 4. What is integrator windup? 5. Why is an antiwindup circuit important? 6. Using the nonlinear saturation function having gain 1 and limits ±1, sketch the block diagram of saturation for an actuator that has gain 7 and limits of ±20. 7. What is a describing function and how is it related to a transfer function? 8. What are the assumptions behind the use of the describing function? 9. What is a limit cycle in a nonlinear system? 10.

How can one determine the describing function for a nonlinear system in the laboratory?

11. What is the minimum time control strategy for a satellite attitude control with bounded controls? 12.

How are the two Lyapunov methods used?

PROBLEMS Problems for Section 9.2: Analysis by Linearization 9.1

Fig. 9.56 shows a simple pendulum system in which a cord is wrapped around a fixed cylinder. The motion of the system that results is described by the differential equation (l + Rθ )θ¨ + g sin θ + Rθ˙ 2 = 0, where l = length of the cord in the vertical (down) position, R = radius of the cylinder. (a) Write the state-variable equations for this system.

Figure 9.56 Motion of cord wrapped around a fixed cylinder

R

u l  Ru l

Problems

651

(b) Linearize the equation around the point θ = 0, and show that for small values of θ, the system equation reduces to an equation for a simple pendulum—that is, θ¨ + (g/l)θ = 0. The circuit shown in Fig. 9.57 has a nonlinear conductance G such that iG = g(vG ) = vG (vG − 1)(vG − 4). The state differential equations are

9.2

di = −i + v, dt dv = −i + g(u − v), dt where i and v are the state variables and u is the input. (a) One equilibrium state occurs when u = 1, yielding i1 = v1 = 0. Find the other two pairs of v and i that will produce equilibrium. (b) Find the linearized model of the system about the equilibrium point u = 1, i1 = v1 = 0. (c) Find the linearized models about the other two equilibrium points.

Figure 9.57 Nonlinear circuit for Problem 9.2

iG

 yG 

 ⵑ

u

L
1

G

C
1



9.3

i

 y 

R
1

Consider the circuit shown in Fig. 9.58; u1 and u2 are voltage and current sources, respectively, and R1 and R2 are nonlinear resistors with the following characteristics: Resistor 1: i1 = G(v1 ) = v13 , Resistor 2: v2 = r(i2 ). y1  i1  R1

Figure 9.58 A nonlinear circuit  u1 (
1 V)

ⵑ 

1F

 x2    x1 

1F

u2 (
27 A)

1H 

R2  y2 

i2
x3

Here the function r is defined in Fig. 9.59. (a) Show that the circuit equations can be written as x˙ 1 = G(u1 − x1 ) + u2 − x3 , x˙ 2 = x3 , x˙ 3 = x1 − x2 − r(x3 ).

652 Chapter 9 Nonlinear Systems y2

Figure 9.59 Nonlinear resistance

Slope
1

1

Slope
Ω

1

i2

0

Suppose we have a constant voltage source of 1 volt at u1 and a constant current source of 27 amps (i.e., u1o = 1, u2o = 27). Find the equilibrium state T  xo = x1o , x2o , x3o for the circuit. For a particular input uo , an equilibrium state of the system is defined to be any constant state vector whose elements satisfy the relation x˙ 1 = x˙ 2 = x˙ 3 = 0. Consequently, any system started in one of its equilibrium states will remain there indefinitely until a different input is applied. (b) Due to disturbances, the initial state (capacitance, voltages, and inductor current) is slightly different from the equilibrium and so are the independent sources; that is, u(t) = uo + δu(t), x(t0 ) = x o (t0 ) + δx(t0 ). Do a small-signal analysis of the network about the equilibrium found in (a), displaying the equations in the form δ x˙ 1 = f11 δx1 + f12 δx2 + f13 δx3 + g1 δu1 + g2 δu2 . (c) Draw the circuit diagram that corresponds to the linearized model. Give the values of the elements. 9.4

Consider the nonlinear system 1

x˙ = −x 2 e− x + sin u,

x(0) = 1.

(a) Assume uo = 0 and solve for x o (t). (b) Find the linearized model about the nominal solution in part (a). 9.5

Linearizing effect of feedback: We have seen that feedback can reduce the sensitivity of the input–output transfer function with respect to changes in the plant transfer function, and reduce the effects of a disturbance acting on the plant. In this problem we explore another beneficial property of feedback: It can make the input–output response more linear than the open-loop response of the plant alone. For simplicity, let us ignore all the dynamics of the plant and assume that the plant is described by the static nonlinearity  u, u ≤ 1, y(t) = u+1 , u > 1. 2

Problems

653

(a) Suppose we use proportional feedback u(t) = r(t) + α(r(t) − y(t)), where α ≥ 0 is the feedback gain. Find an expression for y(t) as a function of r(t) for the closed-loop system. (This function is called the nonlinear characteristic of the system.) Sketch the nonlinear transfer characteristic for α = 0 (which is really open loop), α = 1, and α = 2. (b) Suppose we use integral control: u(t) = r(t) +

t 0

(r(τ ) − y(τ )) dτ .

The closed-loop system is therefore nonlinear and dynamic. Show that if r(t) is a constant, say r, then lim y(t) = r. Thus, the integral control makes the steady-state t→∞

transfer characteristic of the closed-loop system exactly linear. Can the closed-loop system be described by a transfer function from r to y? 9.6

This problem shows that linearization does not always work. Consider the system x˙ = αx 3 ,

x(0) = 0.

(a) Find the equilibrium point and solve for x(t). (b) Assume that α = 1. Is the linearized model a valid representation of the system? (c) Assume that α = −1. Is the linearized model a valid representation of the system? 9.7

Consider the object moving in a straight line with constant velocity shown in Fig. 9.60. The only available measurement is the range to the object. The system equations are ⎤ ⎡ x˙ 0 ⎣ v˙ ⎦ = ⎣ 0 z˙ 0 ⎡

1 0 0

⎤ ⎤⎡ x 0 0 ⎦⎣ v ⎦, z 0

where z = constant, x˙ = constant = v0 ,  r = x 2 + z2 . Derive a linear model for this system.

Figure 9.60 Diagram of moving object for Problem 9.7

z Object

0

x

654 Chapter 9 Nonlinear Systems Problems for Section 9.3: Equivalent Gain Analysis Using the Root Locus 9.8

Consider the third-order system shown in Fig. 9.61. (a) Sketch the root locus for this system with respect to K, showing your calculations for the asymptote angles, departure angles, and so on. (b) Using graphical techniques, locate carefully the point at which the locus crosses the imaginary axis. What is the value of K at that point? (c) Assume that, due to some unknown mechanism, the amplifier output is given by the following saturation non-linearity (instead of by a proportional gain K): ⎧ |e| ≤ 1, ⎨ e, 1, e > 1, u= ⎩ −1, e < −1. Qualitatively describe how you would expect the system to respond to a unit-step input.

Figure 9.61 Control system for Problem 9.8

R







e

K

u

(s  1)2 s3

Y

Problems for Section 9.4: Equivalent Gain Analysis Using Frequency Response: Describing Functions 9.9

Compute the describing function for the relay with dead-zone nonlinearity shown in Fig. 9.6(c).

9.10 Compute the describing function for gain with dead-zone nonlinearity shown in Fig. 9.6(d). 9.11 Compute the describing function for the preloaded spring or Coulomb plus viscous friction nonlinearity shown in Fig. 9.6(e). 9.12 Consider the quantizer function shown in Fig. 9.62 that resembles a staircase. Find the describing function for this nonlinearity and write a MATLAB .m function to generate it. 9.13

Derive the describing function for the ideal contactor controller shown in Fig. 9.63. Is it frequency dependent? Would it be frequency dependent if it had a time delay or hysteresis? Graphically sketch the time histories of the output for several amplitudes of the input and determine the describing function values for those inputs.

9.14 A contactor controller of an inertial platform is shown in Fig. 9.64, where I = 0.1 kg · m2 , I = 10 sec, B h = 1, c J = 0.01 sec, c

Problems Figure 9.62 Quantizer nonlinearity for Problem 9.12

y 4q 3q 2q q h d2

d1

Figure 9.63 Contactor for Problem 9.13

d4

d3

Output T Input

d

Figure 9.64 Block diagram of the system for Problem 9.14

u

Motor and controller

Gimballed

T

1 Is  B

d

v

Gyro K

tLs  1 tf s  1

wm

h cs( Jc s  1)

τL = 0.1 sec, τf = 0.01 sec, d = 10−5 rad, T = 1 N · m.

1 s

w

655

656 Chapter 9 Nonlinear Systems The required stabilization resolution is approximately 10−6 rad: Kϕm > d

for

ϕm > 10−6 rad.

Discuss the existence, amplitude, and frequency of possible limit cycles as a function of the gain K and the DF of the controller. Repeat the problem for a deadband with hysteresis. 9.15 Nonlinear Clegg Integrator: Here there have been some attempts over the years to improve upon the linear integrator. A linear integrator has the disadvantage of having a phase lag of 90◦ at all frequencies. In 1958, J. C. Clegg suggested that we modify the linear integrator to reset its state, x, to zero whenever the input to the integrator, e, crosses zero (i.e., changes sign). The Clegg integrator has the property that it acts like a linear integrator whenever its input and output have the same sign. Otherwise, it resets its output to zero. The Clegg integrator can be described by x(t) = e(t), x(t+) = 0,

if e(t) = 0, if e(t) = 0,

where the latter equation implies that the state of the integrator, x, is reset to zero immediately after e changes sign. It can be implemented with op-amps and diodes. A potential disadvantage of the Clegg integrator is that it may induce oscillations. (a) Sketch the output of the Clegg integrator if the input is e = a sin(ωt). (b) Prove that the DF for the Clegg integrator is N(a, ω) =

4 1 −j . πω ω

and this amounts to a phase lag of only 38◦ .



Problems for Section 9.5: Analysis and Design Based on Stability 9.16 Compute and sketch the optimal reversal curve and optimal control for the minimal time control of the plant x˙ 1 = x2 , x˙ 2 = −x2 + u, |u| ≤ 1. Use the reverse-time method and eliminate the time. 9.17 Sketch the optimal reversal curve for the minimal time control with |u| ≤ 1 of the linear plant x˙ 1 = x2 , x˙ 2 = −2x1 − 3x2 + u. 9.18 Sketch the time-optimal control law for x˙ 1 = x2 , x˙ 2 = −x1 + u, |u| ≤ 1, and show a trajectory for x1 (0) = 3 and x2 (0) = 0.

Problems

657

9.19 Consider the thermal control system shown in Fig. 9.65. The physical plant can be a room, an oven, etc. (a) What is the limit-cycle period? (b) If Tr is commanded as a slowly increasing function, sketch the output of the system, T . Show the solution for Tr “large.”

Figure 9.65 Thermal system for Problem 9.19



T

N

e

 

B sa

T

y

h Sensor 1

9.20 Several systems, such as a spacecraft, a spring-mass system with resonant frequency well below the frequency of switching, a large motor-driven load with very small friction, etc., can be modeled as just an inertia. For an ideal switching curve, sketch the phase portraits of the system. The switching function is e = θ + τ ω. Assume that τ = 10 sec and the control signal = 10−3 rad/sec2 . Now sketch the results with (a) (b) (c) (d)

deadband, deadband plus hysteresis, deadband plus time delay T , deadband plus a constant disturbance.

9.21 Compute the amplitude of the limit cycle in the case of satellite attitude control with delay I θ¨ = N u(t − ), using

u = −sgn(τ θ˙ + θ ).

Sketch the phase-plane trajectory of the limit cycle and time history of θ giving the maximum value of θ . 9.22 Consider the point mass pendulum with zero friction as shown in Fig. 9.66. Using the method of isoclines as a guide, sketch the phase-plane portrait of the motion. Pay particular attention to the vicinity of θ = π. Indicate a trajectory corresponding to spinning of the bob around and around rather than oscillating back and forth.

Figure 9.66 Pendulum for Problem 9.22 u

l

M

658 Chapter 9 Nonlinear Systems 9.23 Draw the phase trajectory for a system x¨ = 10−6 m/sec2 between x˙ (0) = 0, x(0) = 0, and x(t) = 1 mm. Find the transition time tf by graphical means from the parabolic curve by comparing your solution with two different interval sizes and the exact solution. 9.24 Consider the system with equations of motion θ¨ + θ˙ + sin θ = 0. (a) What physical system does this correspond to? (b) Draw the phase portraits for this system. (c) Show a specific trajectory for θ0 = 0.5 rad and θ˙ = 0. 9.25 Consider the nonlinear upright pendulum with a motor at its base as an actuator. Design a feedback controller to stabilize this system. 9.26 Consider the system x˙ = − sin x. Prove that the origin is an asymptotically stable equilibrium point. 9.27 A first-order nonlinear system is described by the equation x˙ = −f (x), where f (x) is a continuous and differentiable nonlinear function that satisfies the following: f (0) = 0, f (x) > 0,

for

x > 0,

f (x) < 0,

for

x < 0.

Use the Lyapunov function V (x) = x 2 /2 to show that the system is stable near the origin

(x = 0).

9.28 Use the Lyapunov equation FT P + PF = −Q = −I to find the range of K for which the system in Fig. 9.67 will be stable. Compare your answer with the stable values for K obtained using Routh’s stability criterion.

Figure 9.67 Control system for Problem 9.28

R(s)
0 

 

K (s  4)(s  1)

9.29 Consider the system d dt



x1 x2

Y(s)



 =

x1 + x2 u x2 (x2 + u)

 ,

y = x1 .

Find all values of α and β for which the input u(t) = αy(t) + β will achieve the goal of maintaining the output y(t) near 1. 9.30 Consider the nonlinear autonomous system ⎤ ⎡ ⎤ ⎡ x (x − x1 ) x d ⎣ 1 ⎦ ⎣ 2 23 ⎦. x2 = x1 − 1 dt x3 −x1 x3

Problems

659

(a) Find the equilibrium point(s). (b) Find the linearized system about each equilibrium point. (c) For each case in part (b), what does Lyapunov theory tell us about the stability of the nonlinear system near the equilibrium point? 9.31 Van der Pol’s equation: Consider the system described by the nonlinear equation x¨ + ε(1 + x 2 )˙x + x = 0 with the constant ε > 0. (a) Show that the equations can be put in the form [Liénard or (x, y) plane]   x3 −x , x˙ = y + ε 3 y˙ = −x. (b) Use the Lyapunov function V = 21 (x 2 + x˙ 2 ), and sketch the region of stability as predicted by this V in the Liénard plane. (c) Plot the trajectories of part (b) and show the initial conditions that tend to the origin. Simulate the system in SIMULINK using various initial conditions on x(0) and x˙ (0). Consider two cases, with ε = 0.5 and ε = 1.0.

10

Control System Design: Principles and Case Studies

A Perspective on Design Principles In Chapters 5, 6, and 7 we presented techniques for analyzing and designing feedback systems based on the root-locus, frequency-response, and statevariable methods. Thus far we have had to consider somewhat isolated, idealized aspects of larger systems and to focus on applying one analysis method at a time. In this chapter we return to the theme of Chapter 4—the advantages of feedback control—to reconsider the overall problem of control systems design with the sophisticated tools developed in Chapters 5 to 7 and 9 in hand. We will apply these tools to several complex, real-world applications in a case study-type format. Having an overarching, step-by-step design approach serves two purposes: It provides a useful starting point for any real-world controls problem, and it provides meaningful checkpoints once the design process is underway. This chapter develops just such a general approach, which will be applied in the case studies.

660

Chapter Overview

661

Chapter Overview Section 10.1 opens the chapter with a step-by-step design process that is sufficiently general to apply to any control design process, but which also provides useful definitions and directions. We then apply the design process to four practical, complex applications: design of an attitude control system for a satellite (Section 10.2), lateral and longitudinal control of a Boeing 747 (Section 10.3), and control of the fuel–air ratio in an automotive engine (Section 10.4), control of a disk drive (Section 10.5), and control of rapid thermal processing (RTP) system (Section 10.6). The satellite case study is representative of the control of geosynchronous communications satellite systems. The study addresses the design of robust control systems in which the physical parameters are known to vary within a given range. In this system the control system needs to meet specifications from the “beginning-of-life” (BOL) to the “end-of-life” (EOL) that spans a period of 12–15 years. The satellite’s moment of inertia and mass will vary as fuel is expended for attitude control, and by deployment and re-orientation of satellite antennas. The satellite case study illustrates the use of a notch compensation for a system with lightly damped resonance. We will see from this case study that collocated actuator and sensor systems are much easier to control than the noncollocated systems. The Boeing 747 case study addresses the familiar flight control system of commercial passenger aircraft. The nonlinear equations of motion are given and are linearized about a particular flight condition. The rigid body dynamics, longitudinal and lateral-directional, are each fourth order. Of course, the flexible modes need also be considered for a more accurate model. The Boeing 747 lateral-stabilization case study will illustrate the use of feedback as an inner-loop designed to aid the pilot, who will provide the primary outer-loop control. The altitude control will show how to combine inner-loop feedback with outer-loop compensation to design a complete control system. The airto-fuel ratio automotive case study is a real-world example that includes a nonlinear sensor and a pure time delay. We will use the describing function method of Chapter 9 to analyze the behavior of this system. Another familiar problem to every PC user is the control of data stored on a disk drive. This case study is about position control and bandwidth will be a key performance parameter. The RTP case study from semiconductor wafer fabrication is remarkably close to the industrial application. The problem concerns temperature tracking and disturbance rejection for a highly nonlinear thermal system. The actuator (lamp) is also nonlinear and we will use the technique from Chapter 9 to try to cancel the effect of this nonlinearity. Another key aspect of this system is actuator saturation and the fact that the control signal cannot go negative. In all these case studies the designer needs to be able to use multiple tools from previous chapters, including the root locus, the frequency response, pole placement by state feedback, and (nonlinear) simulation of time responses to obtain a satisfactory design. In Section 10.7 we present a case study from the emerging field of systems biology and describe chemotaxis or how Escherichia coli (E. coli) swims away from trouble. Section 10.8 provides a historical perspective on applications of feedback control.

662 Chapter 10 Control System Design: Principles and Case Studies

10.1 An Outline of Control Systems Design

Specifications

Sensors

Control engineering is an important part of the design process of many dynamic systems. As suggested in Chapter 4, the deliberate use of feedback can stabilize an otherwise unstable system, reduce the error due to disturbance inputs, reduce the tracking error while following a command input, and reduce the sensitivity of a closed-loop transfer function to small variations in internal system parameters. In those situations for which feedback control is required, it is possible to outline an approach to control systems design that often leads to a satisfactory solution. Before describing this approach, we wish to emphasize that the purpose of control is to aid the product or process—the mechanism, the robot, the chemical plant, the aircraft, or whatever—to do its job. Engineers engaged in other areas of the design process are increasingly taking the contribution of control into account early in their plans. As a result, more and more systems have been designed so that they will not work at all without feedback. This is especially significant in the design of highperformance aircraft, where control has taken its place along with structures and aerodynamics as essential to assuring that the craft will even fly at all. It is impossible to give a description of such overall design in this book, but recognizing the existence of such cases places in perspective not only the specific task of control system design but also the central role this task can play in an enterprise. Control system design begins with a proposed product or process whose satisfactory dynamic performance depends on feedback for stability, disturbance regulation, tracking accuracy, or reduction of the effects of parameter variations. We will give an outline of the design process that is general enough to be useful whether the product is an electronic amplifier or a large structure to be placed in earth orbit. Obviously, to be so widely applicable, our outline has to be vague with respect to physical details and specific only with respect to the feedback-control problem. To present our results, we will divide the control design problem into a sequence of characteristic steps. STEP 1. Understand the process and translate dynamic performance requirements into time, frequency, or pole–zero specifications. The importance of understanding the process, what it is intended to do, how much system error is permissible, how to describe the class of command and disturbance signals to be expected, and what the physical capabilities and limitations are can hardly be overemphasized. Regrettably, in a book such as this, it is easy to view the process as a linear, time-invariant transfer function capable of responding to inputs of arbitrary size, and we tend to overlook the fact that the linear model is a very limited representation of the real system, valid only for small signals, short times, and particular environmental conditions. Do not confuse the approximation with the reality. You must be able to use the simplified model for its intended purpose, and to return to an accurate model or the actual physical system to really verify the design performance. Typical results of this step are specifications that the system have a step response inside some constraint boundaries (as shown in Fig. 10.1a), an open-loop frequency response satisfying certain constraints (Fig. 10.1b), or closed-loop poles to the left of some constraint boundary (Fig. 10.1c). STEP 2. Select sensors. In sensor selection, consider which variables are important to control and which variables can physically be measured. For example, in a jet engine there are critical internal temperatures that must be controlled, but that cannot

663

10.1 An Outline of Control Systems Design

log G

Im(s)

y(t)

vc

log v Re(s)

0 ⬔G 180

t

log v (b)

(a)

(c)

Figure 10.1 Examples of (a) time-response; (b) frequency-response; and (c) pole–zero specifications resulting from Step 10.1 be measured directly in an operational engine. Select sensors that indirectly allow a good estimate to be made of these critical variables. It is important to consider sensors for the disturbance. Sometimes, especially in chemical processes, it is beneficial to sense a load disturbance directly, because improved performance can be obtained if this information is fed forward to the controller. Following are some factors that influence sensor selection: Number of sensors and locations: Technology: Functional performance: Physical properties: Quality factors: Cost:

Actuators

Select minimum required number of sensors and their optimal locations Electric or magnetic, mechanical, electromechanical, electro-optical, piezoelectric Linearity, bias, accuracy, bandwidth, resolution, dynamic range, noise Weight, size, strength Reliability, durability, maintainability Expense, availability, facilities for testing and maintenance

STEP 3. Select actuators. In order to control a dynamic system, obviously you must be able to influence the response. The device that does this is the actuator. Before choosing a specific actuator, consider which variables can be influenced. For example, in a flight vehicle many configurations of movable surfaces are possible, and the influence these have on the performance and controllability of the craft can be profound. The locations of jets or other torque devices are also a major part of the control design of spacecraft. Having selected a particular variable to control, you may need to consider other factors: Number of actuators and locations: Technology: Functional performance: Physical properties: Quality factors: Cost:

Select required actuators and their optimal locations Electric, hydraulic, pneumatic, thermal, other Maximum force possible, extent of the linear range, maximum speed possible, power, efficiency Weight, size, strength Reliability, durability, maintainability Expense, availability, facilities for testing and maintenance

664 Chapter 10 Control System Design: Principles and Case Studies Linearization

STEP 4. Make a linear model. Here you take the best choice for process, actuator, and sensor; identify the equilibrium point of interest; and construct a small-signal dynamic model valid over the range of frequencies included in the specifications of Step 1. You should also validate the model with experimental data where possible. To be able to make use of all the available tools, express the model in state-variable and pole–zero form as well as in frequency-response form. As we have seen, MATLAB
and other computer-aided control systems design software packages have the means to perform the transformations among these forms. Simplify and reduce the order of the model if necessary. Quantify model uncertainty.

Simple compensation PID/lead–lag design

STEP 5. Try a simple proportional–integral–derivative (PID) or lead–lag design. To form an initial estimate of the complexity of the design problem, sketch a frequency response (Bode plot) and a root locus with respect to plant gain. If the plant–actuator– sensor model is stable and minimum phase, the Bode plot will probably be the most useful; otherwise, the root locus shows very important information with respect to behavior in the right half-plane (RHP). In any case, try to meet the specifications with a simple controller of the lead–lag variety, including integral control if steady-state error response requires it. Do not overlook feed-forward of the disturbances if the necessary sensor information is available. Consider the effect of sensor noise, and compare a lead network to a direct sensor of velocity to see which gives a better design. STEP 6. Evaluate/modify plant. Based on the simple control design, evaluate the source of the undesirable characteristics of the system performance. Reevaluate the specifications, the physical configuration of the process, and the actuator and sensor selections in light of the preliminary design, and return to Step 1 if improvement seems necessary or feasible. For example, in many motion-control problems, after testing the first-pass design, you might find vibrational modes that prevent the design from meeting the initial specifications of the problem. It may be much easier to meet the specifications by altering the structure of the plant through the addition of stiffening members or by passive damping than to meet them by control strategies alone. An alternative solution may be to move a sensor so it is at a node of a vibration mode, thus providing no feedback of the motion. Also, some actuator technologies (such as hydraulic) have many more low-frequency vibrations than others (such as electric) do and changing the actuator technology may be indicated. In a digital implementation, it may be possible to revise the sensor–controller–actuator system structure so as to reduce time delay, which is always a destabilizing element. In thermal systems, it is often possible to change heat capacity or conductivities by material substitution that will enhance the control design. It is important to consider all parts of the design, not only the control logic, to meet the specifications in the most cost-effective way. If the plant is modified, go back to Step 1. If the design now seems satisfactory, go to Step 8; otherwise try Step 7.

Optimal design

STEP 7. Try an optimal design. If the trial-and-error compensators do not give entirely satisfactory performance, consider a design based on optimal control. The symmetric root locus (SRL) will show possible root locations from which to select locations for the control poles that meet the response specifications; you can select locations for the estimator poles that represent a compromise between sensor and process noise. Plot the corresponding open-loop frequency response and the root locus to evaluate

10.1 An Outline of Control Systems Design

665

the stability margins of this design and its robustness to parameter changes. You can modify the pole locations until a best compromise results. Returning to the SRL with different cost measures is often a part of this step, or computations via the direct functions lqr and lqe can be used. Another variation on optimal control is to propose a fixed structure controller with unknown parameters, formulate a performance cost function, and use parameter optimization to find a good set of parameter values. Compare the optimal design yielding the most satisfactory frequency response with the transform-method design you derived in Step 5. Select the better of the two before proceeding to Step 8. STEP 8. Build a computer model, and compute (simulate) the performance of the design. After reaching the best compromise among process modification, actuator and sensor selection, and controller design choice, run a computer model of the system. This model should include important nonlinearities, such as actuator saturation, realistic noise sources, and parameter variations you expect to find during operation of the system. The simulation will often identify sensitivities that may lead to going back to Step 5 or even Step 2. Design iterations should continue until the simulation confirms acceptable stability and robustness. As part of this simulation you can often include parameter optimization, in which the computer tunes the free parameters for best performance. In the early stages of design the model you simulate will be relatively simple; as the design progresses, you will study more complete and detailed models. At this step it is also possible to compute a digital equivalent of the analog controller as described in Chapters 4 and 8. Some refinement of the controller parameters may be required to account for the effects of digitization. This allows the final design to be implemented with digital processor logic. If the results of the simulation prove the design satisfactory, go to Step 9, otherwise return to Step 1. Prototype Prototype testing

STEP 9. Build a prototype. As the final test before production, it is common to build and test a prototype. At this point you verify the quality of the model, discover unsuspected vibration and other modes, and consider ways to improve the design. Implement the controller using embedded software/hardware. Tune the controller if necessary. After these tests, you may want to reconsider the sensor, actuator, and process and return to Step 1—unless time, money, or ideas have run out. This outline is an approximation of good practice; other engineers will have variations on these themes. In some cases you may wish to carry out the steps in a different order, to omit a step, or to add one. The stages of simulation and prototype construction vary widely, depending on the nature of the system. For systems for which a prototype is difficult to test and rework (for example, a satellite) or where a failure is dangerous (for example, active stabilization of a high-speed centrifuge or landing a human on the moon), most of the design verification is done through simulation of some sort. It may take the form of a digital numerical simulation, a laboratory scale model, or a full-size laboratory model with a simulated environment. For systems that are easy to build and modify (for example, feedback control for an automotive fuel system), the simulation step is often skipped entirely; design verification and refinement are instead accomplished by working with prototypes.

666 Chapter 10 Control System Design: Principles and Case Studies One of the issues raised in the preceding discussion (Step 6) was the important consideration for changing the plant itself. In many cases, proper plant modifications can provide additional damping or increase in stiffness, change in mode shapes, reduction of system response to disturbances, reduction of Coulomb friction, change in thermal capacity or conductivity, etc. It is worth elaborating on this by way of specific examples from the authors’ experiences. In a semiconductor wafer-processing example, the edge ring holding the wafer was identified as a limiting factor in closedloop control. Modifying the thickness of the edge ring and using a different coating material reduced the heat losses and, together with relocating one of the temperature sensors closer to the edge ring, resulted in significant improvement in control performance. In another application, thin film processing, simply changing the order of the two incoming flows resulted in significant improvement in the mixing of the precursor and oxidizer materials, and led to improvement in uniformity of the film. In an application on physical vapor deposition using RF-plasma, the shape of the target was modified to be curved to counter the geometry effects of the chamber, and yielded substantial improvements in deposition uniformity. As the last example, in a hydraulic spindle control problem, adding oil temperature control with ceramic insulation and a temperature sink for the bell housing resulted in several orders of magnitude reduction in disturbances not achievable by feedback control alone.1 One can also mention aerospace applications for which the control was an afterthought, and the feedback control problem became exceedingly difficult and resulted in poor closed-loop performance. The moral of this discussion is that one must not forget the option of modifying the plant itself to make the control problem easier and provide maximum closed-loop performance. The usual approach of designing the system and “throwing it over the fence” to the control group has proved to be inefficient and flawed. A better approach that is gaining momentum is to get the control engineer involved from the onset of a project to provide early feedback on how hard it is to control the system. The control engineer can provide valuable feedback on choice of actuators and sensors and can even suggest modifications to the plant. It is often much more efficient to change the plant design while it is on the drawing board before “any metal has been bent.” Closed-loop performance studies can then be performed on a simple model of the system early on. Implicit in the process of design is the well-known fact that designs within a given category often draw on experience gained from earlier models. Thus, good designs evolve rather than appear in their best form after the first pass. We will illustrate the method with several cases (Sections 10.2 to 10.6). For easy reference, we summarize the steps here.

Summary of Control Design Steps 1. Understand the process and its performance requirements. 2. Select the types and number of sensors considering location, technology, and noise.

1 Our colleague Prof. Daniel DeBra strongly believes in considering modifying the plant itself an option

for improved control. He cites this particular application to make the point. Of course, we agree with him!

10.2 Design of a Satellite’s Attitude Control

667

3. Select the types and number of actuators considering location, technology, noise, and power. 4. Make a linear model of the process, actuator, and sensor. 5. Make a simple trial design based on the concepts of lead–lag compensation or PID control. If satisfied, go to Step 8. 6. Consider modifying the plant itself for improved closed-loop control. 7. Make a trial pole-placement design based on optimal control or other criteria. 8. Simulate the design, including the effects of nonlinearities, noise, and parameter variations. If the performance is not satisfactory, return to Step 1 and repeat. Consider modifying the plant itself for improved closed-loop control. 9. Build a prototype and test it. If not satisfied, return to Step 1 and repeat.

10.2 Design of a Satellite’s Attitude Control Our first example, taken from the space program, is suggested by the need to control the pointing direction, or attitude, of a satellite in orbit about the earth. Figure 10.2(a) shows a picture of a geosynchronous communications satellite. We will go through each step in our design outline and touch on some of the factors that might be considered for the control of such a system. STEP 1. Understand the process and its performance specifications. A satellite is sketched in Fig. 10.2(b). We imagine that the vehicle has an astronomical survey mission requiring accurate pointing of a scientific sensor package. This package must be maintained in the quietest possible environment, which entails isolating it Figure 10.2 (a) Picture of the geosynchronous communications satellite IPSTAR; (b) diagram of a satellite and its two-body model Source: Courtesy Thaicom plc and Space Systems/Loral

668 Chapter 10 Control System Design: Principles and Case Studies from the vibrations and electrical noise of the main service body and from its power supplies, thrusters, and communication gear. We model the resulting structure as two masses connected by a flexible boom. In Fig. 10.2(b), the satellite attitude θ2 is the angle between the star sensor and the instrument package, and θ1 is the angle of the main satellite with respect to the star. Figure 10.2(b) shows the equivalent mechanical system diagram for the satellite, where the sensor is mounted to the disk associated with θ2 . Disturbance torques due to solar pressure, micrometeorites, and orbit perturbations are computed to be negligible. The pointing requirement arises when it is necessary to point the unit in another direction. It can be met by dynamics with a transient settling time of 20 sec and an overshoot of no more than 15%. The dynamics of the satellite include parameters that can vary. The control must be satisfactory for any parameter values in a prespecified range to be given when the equations are written. STEP 2. Select sensors. In order to orient the scientific package, it is necessary to measure the attitude angles of the package. For this purpose we propose to use a star tracker, a system based on gathering an image of a specific star and keeping it centered on the focal plane of a telescope. This sensor gives a relatively noisy but very accurate (on the average) reading proportional to θ2 , the angle of deviation of the instrument package from the desired angle. To stabilize the control, we include a rate gyro to give a clean reading of θ˙2 , because a lead network on the star-tracker signal would amplify the noise too much. Furthermore, the rate gyro can stabilize large motions before the star tracker has acquired the target star image. STEP 3. Select actuators. Major considerations in selecting the actuator are precision, reliability, weight, power requirements, and lifetime. Alternatives for applying torque are cold-gas jets, reaction wheels or gyros, magnetic torquers, and a gravity gradient. The jets have the most power and are the least accurate. Reaction wheels are precise but can transfer only momentum, so jets or magnetic torquers are required to “dump” momentum from time to time. Magnetic torquers provide relatively low levels of torque and are suitable only for some low-altitude satellite missions. A gravity gradient also provides a very small torque that limits the speed of response and places severe restrictions on the shape of the satellite. For purposes of this mission, we select cold-gas jets as being fast and adequately accurate. STEP 4. Make a linear model. For the satellite, we assume two masses connected by a spring with torque constant k and viscous-damping constant b as shown in Fig. 10.2. The equations of motion are J1 θ¨1 + b(θ˙1 − θ˙2 ) + k(θ1 − θ2 ) = Tc , J2 θ¨2 + b(θ˙2 − θ˙1 ) + k(θ2 − θ1 ) = 0,

(10.1a) (10.1b)

where Tc is the control torque on the main body. With inertias J1 = 1 and J2 = 0.1, the transfer function is G(s) =

10bs + 10k . + 11bs + 11k)

s2 (s2

(10.2)

10.2 Design of a Satellite’s Attitude Control If we choose x = [ θ2

θ˙2

θ1

669

θ˙1 ]T

as the state vector, then, using Eq. (10.1a) and assuming Tc ≡ u, we find that the equations of motion in state-variable form are ⎡ ⎤ ⎤ ⎡ 0 1 0 0 0 k k b ⎥ b ⎢ ⎥ ⎢ ⎢ 0 ⎥ ⎥ ⎢ − − ⎢ ⎥ ⎥ ⎢ J2 J J J 2 2 2 ⎥x + ⎢ (10.3a) x˙ = ⎢ ⎥ u, 0 0 1 ⎥ ⎢ 0 ⎥ ⎢ 0 ⎣ 1 ⎦ ⎣ k b k b ⎦ − − J1 J1 J1 J1 J1  y = 1 0 0 0 x. (10.3b) Physical analysis of the boom leads us to assume that the parameters k and b vary as a result of temperature fluctuations but are bounded by

Selecting nominal values for varying parameters

0.09 ≤ k ≤ 0.4, (10.4a)   k k 0.038 ≤ b ≤ 0.2 . (10.4b) 10 10 As a result, the vehicle’s natural resonance frequency ωn can vary between 1 and 2 rad/sec, and the damping ratio ζ varies between 0.02 and 0.1. One approach to control design when parameters are subject to variation is to select nominal values for the parameters, construct the design for this model, and then test the controller performance with other parameter values. In the present case we choose nominal values of ωn = 1 and ζ = 0.02. The choice is somewhat arbitrary, being based on experience and heuristic analysis. However, note that these are the lowest values in their respective ranges and thus correspond to the plant that is probably the most difficult to control so as to meet the specifications. We assume that a design for this model has a good chance to meet the specifications for other parameter values as well. (Another choice would be to select a model with average values for each parameter.) The selected parameter values are k = 0.091 and b = 0.0036; with J1 = 1 and J2 = 0.1, the nominal equations become ⎤ ⎡ ⎤ ⎡ 0 1 0 0 0 ⎥ ⎢ −0.91 −0.036 ⎢ 0.91 0.036 ⎥ ⎥ x + ⎢ 0 ⎥ u, (10.5a) x˙ = ⎢ ⎣ ⎦ ⎣ 0 ⎦ 0 0 0 1 1 0.091 0.0036 −0.091 −0.0036 y=[ 1

0

0

0 ]x.

(10.5b)

The corresponding transfer function, using the MATLAB ss2tf function, is then G(s) =

0.036(s + 25) . + 0.04s + 1)

s2 (s2

(10.6)

When a trial design is completed, the computer simulation should be run with a range of possible parameter values to ensure that the design has adequate robustness to tolerate these changes. Equations (3.66)–(3.68) tell us that the dynamic performance specifications will be met if the closed-loop poles have a natural frequency of 0.5 rad/sec and a closed-loop damping ratio of 0.5; these correspond to an open-loop

670 Chapter 10 Control System Design: Principles and Case Studies Figure 10.3 Root locus of KG(s)

2 1.5 Imaginary axis

1 0.5 0 0.5 1 1.5 2 2 1.5 1 0.5

0

0.5

1

1.5

2

40

10

20

db

100

0

1

20

0.1 0.01 0.1

1

40 10

v (rad/sec) (a)

Phase

Figure 10.4 Open-loop Bode plot of KG(s) for K = 0.5

Magnitude, KG(s)

Real axis

180 200 220 240 260 280 300 320 340 360 0.1

1 v (rad/sec)

10

(b)

crossover frequency of ωc ∼ = 0.5 rad/sec and a phase margin of about PM = 50◦ . We will try to meet these design criteria. STEP 5. Try a lead–lag or PID controller. The proportional-gain root locus for the nominal plant is drawn in Fig. 10.3, and the Bode plot is given in Fig. 10.4. We can see

10.2 Design of a Satellite’s Attitude Control Figure 10.5 Root locus of KD1 (s)G(s)

671

Im(s) 2.0

2.0

1.0

1.0

2.0 Re(s)

2.0

from Fig. 10.4 that this may be a difficult design problem because the frequency of the lightly damped resonance is greater than the crossover-frequency design point by only a factor of 2. This situation will require that the compensation can correct for the phase lag of the plant at the resonance frequency. Such correction is very dependent on knowing the resonance frequency, which is subject to change in this case. There may be trouble ahead. In order to illustrate some important aspects of compensation design, we will at first ignore the resonance and generate a design that would be acceptable for the rigid body alone. We take the process transfer function to be 1/s2 , the feedback to be position plus derivative (star tracker plus rate gyro) or PD control with the transfer function D(s) = K(sTD + 1), and the response objective to be ωn = 0.5 rad/sec and ζ = 0.5. A suitable controller would be D1 (s) = 0.25(2s + 1).

(10.7)

The root locus for the actual plant with D1 is shown in Fig. 10.5 and the Bode plot in Fig. 10.6. From these plots we can see that the low-frequency poles are reasonable but that the system will be unstable because of the resonance.2 At this point we take the simple actions of reducing our expectation with respect to bandwidth and of slowing the system down by lowering the gain until the system is stable. With so little damping, we must really go slowly. A bit of experimentation leads to D2 (s) = 0.001(30s + 1),

(10.8)

for which the root locus is drawn in Fig. 10.7 and the Bode plot given in Fig. 10.8. The Bode plot shows that we have a phase margin of 50◦ but a crossover frequency of only ωc = 0.04 rad/sec. While this is too low to meet the settling-time specification, a low crossover frequency is unavoidable if we expect to keep the gain at the resonance frequency below unity so that it is gain stabilized. An alternative approach to the problem is to place zeros near the lightly damped poles and use them to hold these poles back from the RHP. Such a compensation has 2 If this system were built, the actuator jets would saturate as the response grew. We could analyze the response using the method described in Section 9.3 for nonlinear systems. From the analysis we would expect the signal to grow and the equivalent gain of the actuator to fall until the roots return to the imaginary axis near ωn . The resulting limit cycle would rapidly deplete the control gas supply.

100

40

10

20 0

1 0.1

20

0.01

40

0.001

60

0.0001 0.1

1 v (rad/sec)

80 10

Phase

(a) 120 140 160 180 200 220 240 260 280 300 0.1

1 v (rad/sec) (b)

Figure 10.7 Root locus of KD2 (s)G(s)

Im(s)

0.5

0.5

Re(s)

0.5

10

db

Figure 10.6 Bode plot of KD1 (s)G(s)

Magnitude, KD1(s)G(s)

672 Chapter 10 Control System Design: Principles and Case Studies

10.2 Design of a Satellite’s Attitude Control 20

Magnitude, D2(s)G(s)

10

0

1

20

0.1

0.01 0.01

0.1

1

10

1

10

db

Figure 10.8 Bode plot of D2 (s)G(s)

673

40

v (rad/sec)

Phase

(a)

80 100 120 140 160 180 200 220 240 260 280 0.01

0.1 v (rad/sec) (b)

Figure 10.9 Twin-tee realization of a notch filter





ei

eo



 Notch

Notch filter

a frequency response with a very low gain near the frequency of the offending poles and a reasonable gain elsewhere. Because the frequency response seems to have a dent or notch in it, the device is called a notch filter. (It is also called a band reject filter in electric network theory.) An RC circuit with a notch characteristic is shown in Fig. 10.9, its pole–zero pattern in Fig. 10.10, and its frequency response in Fig. 10.11. The +180◦ phase lead of the notch can be used to correct for the 180◦ phase lag of the resonance; if the notch frequency is lower than the plant’s resonance frequency, the system phase is kept above 180◦ near resonance.

674 Chapter 10 Control System Design: Principles and Case Studies Figure 10.10 Notch filter pole–zero pattern

Im(s) 0.9

Re(s)

25

Magnitude

Figure 10.11 Bode plot of a notch filter

1000

60

100

40

10

20 0

1

20

0.1 0.01 0.1

db

0.9

1 v (rad/sec)

40 10

Phase

(a) 180 160 140 120 100 80 60 40 20 0 20 0.1

1 v (rad/sec)

10

(b)

With this idea we return to the compensation given by Eq. (10.7) and add the notch, producing the revised compensator transfer function (s/0.9)2 + 1 . (10.9) [(s/25) + 1]2 The Bode plot for this case is shown in Fig. 10.12, the root locus in Fig. 10.13, and the unit step response in Fig. 10.14. The settling time of the design is too long for the D3 (s) = 0.25(2s + 1)

Magnitude, KD3(s)G(s)

Figure 10.12 Bode plot of KD3 (s)G(s)

100

40

10

20 0

1

20

0.1 0.01 0.1

675

db

10.2 Design of a Satellite’s Attitude Control

vc

1 v (rad/sec)

40 10

(a) 60

Phase

10 40 90 140 190 0.1

PM 1 vc v (rad/sec)

10

(b)

specification and the overshoot is too large, but this design approach seems promising; with iteration it could lead to a satisfactory compensator. We now recall that the compensator is expected to provide adequate performance as the parameters vary over the ranges given by Eq. (10.3a). An examination of the robustness of the design can be made by looking at the root locus shown in Fig. 10.15, which is drawn using the compensator of Eq. (10.9) and the plant with ωn = 2, rather than 1, such that ˆ G(s) =

(s/50 + 1) s2 (s2 /4 + 0.02s

+ 1)

.

(10.10)

This assumes that the boom is as stiff as possible. Notice that now the low-frequency poles have a damping ratio of only 0.02. Combining the various parameter values, we get the frequency response and transient response shown in Figs. 10.16 and 10.17. We could make a few more trial-and-error iterations with the notch filter and rate feedback, but the system is complex enough that a look at state-space designs now seems reasonable. We go to Step 7. STEP 6. Evaluate/modify plant. Refer to the collocated control discussion after Step 8.

676 Chapter 10 Control System Design: Principles and Case Studies Figure 10.13 Root locus of KD3 (s)G(s)

Im(s) 2.0

2.0

1.0

2.0 Re(s)

2.0

Figure 10.14 Closed-loop step response of D3 (s)G(s) where θ2 (0) = 0.2 rad

1.4 1.2

Amplitude

1.0 0.8 0.6 0.4 0.2 0 0

5

10

15 20 25 Time (sec)

Figure 10.15 ˆ Root locus of KD3 (s)G(s)

30

35

Im(s) 3.0

1.0

7.0

4.0 3.0

2.0 1.0 1.0

3.0

Re(s)

40

ˆ Magnitude, KD3(s)G(s)

Figure 10.16 ˆ Bode plot of KD3 (s)G(s)

100

40

10

20 0

1

20

0.1 0.01 0.1

40 10

1 v (rad/sec) (a)

100 50

Phase

0 50 100 150 180 200 0.1

1 v (rad/sec)

10

(b)

Figure 10.17 Closed-loop step ˆ response of D3 (s)G(s)

1.4 1.2

Amplitude

1.0 0.8 0.6 0.4 0.2 0 0

5

10

15 20 25 Time (sec)

30

35

40

db

10.2 Design of a Satellite’s Attitude Control

677

678 Chapter 10 Control System Design: Principles and Case Studies STEP 7. Try an optimal design using pole placement. Using the state-variable formulation of the equations of motion in Eq. (10.4a), we devise a controller that will place the closed-loop poles in arbitrary locations. Of course, used without thought, the method of pole placement can also result in a design that requires unreasonable levels of control effort or is very sensitive to changes in the plant transfer function. Guidelines for pole placement are given in Chapter 7; an often successful approach is to derive optimal pole locations using the SRL. Figure 10.18 shows the SRL for the problem at hand. To obtain a bandwidth of about 0.5 rad/sec, we select closed-loop control poles from this locus at −0.45 ± 0.34j and −0.15 ± 1.05j. If we select αc (s), as discussed earlier, from the SRL, the control gain using the MATLAB function place is  K = −0.2788 0.0546 0.6814 1.1655 . (10.11) Figure 10.19 shows the step responses for the nominal plant parameters and stiffspring plant models. The Bode plot of the SRL controller design with the nominal plant parameters can be computed from the loop transfer function (by breaking the Figure 10.18 Symmetric root locus of the satellite

Im(s) 2.0

2.0

1.0

1.0

2.0 Re(s)

2.0

Figure 10.19 Closed-loop step response of the SRL design

1.2

Nominal

Amplitude

1.0 0.8 Stiff spring 0.6 0.4 0.2 0 0

5

10 Time (sec)

15

20

10.2 Design of a Satellite’s Attitude Control

679

loop at u) KX(s) = K(sI − F)−1 G U(s) and results in a phase margin of about 60◦ , as shown in Fig. 10.20. While the speed of response of the design meets the specifications with the nominal plant, the settling time when the plant has the stiff spring is a bit longer than the specifications call for. We might be able to get a better compromise between the nominal and the stiffspring cases by selecting another point on the SRL; at this point we do not know. The designer must face alternatives such as these and select the best compromise for the problem at hand. The design of Fig. 10.19 is based on full-state feedback. To complete the optimal design, we need an estimator. We select the closed-loop estimator error poles to be about eight times faster than the control poles. The reason for this is to keep the error poles from reducing the robustness of the design; a fast estimator will have almost the same effect on the response as no estimator at all. We choose the error poles from the SRL at −7.7 ± 3.12j and −3.32 ± 7.85j. Pole placement with these values leads

40

20

db

10



Magnitude, KX(jv) U(jv)



100

0

1

0.1 0.1

1 v (rad/sec)

20 10

(a)

40 60 80 Phase

Figure 10.20 Frequency response of the SRL design from u to Kx

100 120 140

PM

160 180 0.1

1 v (rad/sec) (b)

10

680 Chapter 10 Control System Design: Principles and Case Studies to an estimator (filter) gain, using the MATLAB function place: ⎡ ⎤ 22 ⎢ 242.3 ⎥ ⎥ L=⎢ ⎣ 1515.4 ⎦. 5503.9

(10.12)

After we combine the control gain and estimator, as described in Section 7.8, the compensator transfer function that results from Eq. (7.177) is D4 (s) =

−745(s + 0.3217)(s + 0.0996 ± 0.9137 j) . (s + 3.1195 ± 8.3438 j)(s + 8.4905 ± 3.6333 j)

(10.13)

The frequency response of this compensator (Fig. 10.21) shows that pole placement has introduced a notch directly. The frequency response and the root locus of the combined system D4 (s)G(s) are given in Figs. 10.22 and 10.23, while Fig. 10.24 shows the step response for both the nominal and the stiff-spring plants. Notice that the design almost meets the specifications.

40

100

20 0

10

20

1

1 v (rad/sec)

40 10

(a)

250 200 150 100 50 0 0.1

1 v (rad/sec) (b)

10

db

1000

0.1 0.1

Phase

Figure 10.21 Bode plot of optimal compensator D4 (s)

Magnitude, D4(s)

STEP 8. Simulate the design, and compare the alternatives. At this point we have two designs, with differing complexities and different robustness properties. The

100

40

10

20

681

db

Figure 10.22 Bode plot of the compensated system D4 (s)G(s)

Magnitude, D4(s)G(s)

10.2 Design of a Satellite’s Attitude Control

0

1

20

0.1 0.01 0.1

40 10

1 v (rad/sec) (a)

0 50 Phase

100 150 180 200 250 300 0.1

1 v (rad/sec)

10

(b)

Figure 10.23 Root locus of D4 (s)G(s)

Im(s) 2.0

2.0

1.0

1.0

2.0 Re(s)

2.0

notch-filter design might be improved with further iterations or by starting with a different nominal case. The SRL design meets the specifications for the nominal plant but is too slow for the stiff-spring case, although alternative selections for the pole locations might lead to a better design. In either case, much more extensive studies

682 Chapter 10 Control System Design: Principles and Case Studies Figure 10.24 Closed-loop step response of D4 (s)G(s)

1.2

Stiff

Amplitude

1.0 Nominal

0.8 0.6 0.4 0.2 0 0

Collocated actuator and sensor

5

10

15 20 Time (sec)

25

30

need to be made to explore the robustness and noise-response properties. Rather than follow any of these paths, we consider some aspects of the physical system. Both designs are strongly influenced by the presence of the lightly damped resonant mode caused by the coupled masses. However, the transfer function of this system is strongly dependent on the fact that the actuator is on one body and the sensor is on the other (that is, not collocated). Suppose that, rather than considering pointing the star tracker on the small mass, we have the mission of pointing the main mass, perhaps toward an Earth station for communications purposes. For this purpose we can put the sensor on the same mass that holds the actuator—to give control with a collocated actuator and sensor. Due to the physics of the situation, the system’s transfer function now has zeros close to the flexible modes, so control can be achieved by using PD feedback alone, because the plant already has the effect of a notch compensator. Consider the transfer function of the satellite with collocated actuator and sensor (to measure θ1 ) for which the state matrices are ⎡ ⎤ ⎡ ⎤ 0 1 0 0 0 ⎢ −0.91 −0.036 ⎢ ⎥ 0.91 0.036 ⎥ ⎥, G = ⎢ 0 ⎥, F=⎢ ⎣ ⎦ ⎣ 0 ⎦ 0 0 0 1 0.091 0.0036 −0.091 −0.0036 1 H=[ 0

0

1

0 ].

The transfer function of the system using the MATLAB ss2tf function is (s + 0.018 ± 0.954j) . (10.14) Gco (s) = H(sI − F)−1 G = s2 (s + 0.02 ± j) Notice the presence of the zeros in the vicinity of the complex conjugate poles. If we now use the same PD feedback as before, namely, D5 (s) = 0.25(2s + 1),

(10.15)

then the system will not only be stabilized, but will also have a satisfactory response (if we consider θ1 as the output), because the resonant poles tend to be cancelled by the complex conjugate zeros. Figures 10.25–10.27 show the frequency response, the root locus, and the step response, respectively, for this system. Note from Fig. 10.27 that the step response

Magnitude, D5(s)Gco(s)

Figure 10.25 Bode plot of D5 (s)Gco (s)

100

40

10

20 0

1

20

0.1 0.01 0.1

40 10

1 v (rad/sec) (a)

40 60

Phase

80 100 120 140 160 180 0.1

1 v (rad/sec)

10

(b)

Figure 10.26 Root locus for D5 (s)Gco (s)

Im(s) 2.0

2.0

1.0

2.0

2.0 Re(s)

db

10.2 Design of a Satellite’s Attitude Control

683

684 Chapter 10 Control System Design: Principles and Case Studies 1.4 with nominal plant, Gco

1.2 1.0 Amplitude

Figure 10.27 Closed-loop step response of the system with collocated control, D5 (s)Gco (s) and D5 (s)Gˆ co (s)

0.8

ˆ co with perturbed plant, G

0.6 0.4 0.2 0 0

10

20

30

40

50

80

100

Time (sec)

Amplitude

Figure 10.28 Response at θ2 of the collocated design

2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 0

20

40 60 Time (sec)

has the excess overshoot associated with the zero of the compensator in the forward path of the transfer function. The result is a very simple robust design achieved by moving the sensor from a noncollocated position to one collocated with the actuator. The result illustrates that, to achieve good feedback control, it is very important to consider sensor location and other features of the physical problem. However, this last control design will not do for pointing the star tracker. This is evident from plotting the output θ2 corresponding to the nice step response of Fig. 10.27. The result is shown in Fig. 10.28. An architecture suggested by the results is to place a coarse star tracker on the satellite body to be used for search and initial settling. Then switch to a star tracker on the instrument package with longer settling time for fine control.

10.3 Lateral and Longitudinal Control of a Boeing 747 The Boeing 747 (Fig. 10.29) is a large wide-body transport jet. A schematic with the relevant coordinates that move with the airplane is shown in Fig. 10.30. The linearized

10.3 Lateral and Longitudinal Control of a Boeing 747

685

Figure 10.29 Boeing 747 Source: Courtesy Boeing Commercial Airplane Co.

Figure 10.30 Definition of aircraft coordinates

Velocity vector b x, u

u, q

a

y, y Rudder dr

f, p

Elevators de

c, r Aileron da z, w x, y, z
position coordinates u, y, w
velocity coordinates p
roll rate q
pitch rate r
yaw rate

f
roll angle u
pitch angle c
yaw angle b
side-slip angle a
angle of attack

equations of (rigid-body) motion3 for the Boeing 747 are of eighth order but are separated into two fourth-order sets representing the perturbations in longitudinal (U, W , θ , and q in Fig. 10.30) and lateral (φ, β, r, and p) motion. The longitudinal motion consists of axial (X), vertical (Z), and pitching (θ, q) motion, while the lateral motion consists of rolling (φ, p) and yawing (r, β) movement. The side-slip angle β is a measure of the direction of forward velocity relative to the direction of the nose of the airplane. The elevator control surfaces and the throttle affect the longitudinal motion, whereas the aileron and rudder primarily affect lateral motion. Although there is a small amount of coupling of lateral motion into longitudinal motion, this is usually ignored, so the equations of motion are treated as two decoupled fourth-order sets for designing the control, or stability augmentation, for the aircraft.

3 For derivation of equations of motion for an aircraft, the reader is referred to Bryson (1994), Etkin and Reid (1996), and McRuer et al. (1973).

686 Chapter 10 Control System Design: Principles and Case Studies The nonlinear rigid body equations of motion in body-axis coordinates, under proper assumptions,4 can be derived as (Bryson, 1994) m(U˙ + qW − rV ) = X − mg sin θ + κT cos θ, (10.16) m(V˙ + rU − pW ) = Y + mg cos θ sin φ, ˙ + pV − qU) = Z + mg cos θ cos φ − κT sin θ, m(W Ix p˙ + Ixz r˙ + (Iz − Iy )qr + Ixz qp = L,

(10.17)

Iy q˙ + (Ix − Iz )pr + Ixz (r 2 − p2 ) = M, Iz r˙ + Ixz p˙ + (Iy − Ix )qp − Ixz qr = N, where m = mass of the aircraft, [U, V , W ] = body-axis components of the velocity of the center of mass (c.m.),   −1 V , β = tan U [Uo , Vo , Wo ] = reference velocities, [p, q, r] = the body-axis components of the angular velocity of the aircraft (roll, pitch, and yaw respectively) [X, Y , Z] = the body-axis aerodynamic forces about the c.m., [L, M, N] = the body-axis aerodynamic torques about the c.m., go = the gravitational force per unit mass, Ii = the inertias in body axes, (θ, φ) = the Euler pitch and roll angles of the aircraft body axes with respect to horizontal, Vref = reference flight speed, T = the propulsive thrust resultant, and κ = the angle between thrust and body x-axis. The linearization of these equations can be carried out as follows: In the steady-state ˙ = p˙ = q˙ = r˙ = 0. straight, level, and constant speed flight condition, U˙ = V˙ = W Furthermore, there is no turning in any axis so that po = qo = ro = 0, and the wings will be level so that φ = 0. However, there will be an angle of attack in order to provide some lift from the wings to counteract the aircraft’s weight, so θo and Wo = 0, where U = Uo + u, V = Vo + v, W = Wo + w. 4 x–z is the body-axis plane of mass symmetry.

(10.18)

10.3 Lateral and Longitudinal Control of a Boeing 747

687

The steady-state velocity body axis components will be Uo = Vref cos(θo ),

(10.19)

Vo = 0 (βo = 0), Wo = Vref sin(θo ), as depicted in Fig. 10.31. With these conditions, the equilibrium (see Chapter 9) equations are 0 = X0 − mgo sin θ0 + κT cos θ0 ,

(10.20)

0 = Y0 , 0 = Z0 + mgo cos θ0 − κT sin θ0 , 0 = L0 , 0 = M0 , 0 = N0 . Figure 10.31 Steady-state flight condition b Relative wind dr

(a) x a Relative wind

Uo uo Vref

Wo

da

z (b)

f

(c)

de

688 Chapter 10 Control System Design: Principles and Case Studies With the assumptions (Bryson, 1994) (v2 , w2 ) u2 ,

(10.21)

(φ , θ ) 1, 2

2

u2 , b2 where b denotes the wingspan, many of the nonlinear terms in Eqs. (10.16) and (10.17) can be neglected. Substitution of Eq. (10.20) in the nonlinear equations of motion leads to a set of linear perturbational equations that describe small deviations from constant speed, straight and level flight. The equations of motion then divide into two uncoupled sets of longitudinal and lateral equations of motion. For linearized longitudinal motion, the results are ⎤ ⎡ ⎡ ⎤⎡ ⎤ ⎤ ⎡ u˙ Xu Xw −Wo −go cos θo u Xδe ⎢ w˙ ⎥ ⎢ Zu Zw ⎥ ⎢ ⎥ ⎢ Uo −go sin θo ⎥ ⎥ ⎢ ⎢ ⎥ ⎢ w ⎥ + ⎢ Zδe ⎥ δe, (10.22) ⎣ q˙ ⎦ = ⎣ Mu Mw Mq ⎦ ⎣ q ⎦ ⎣ Mδe ⎦ 0 0 0 0 1 0 θ θ˙ (p2 , q2 , r 2 )

where u = forward velocity perturbation in the aircraft in x direction (Fig. 10.30), w = velocity perturbation in the z direction (also proportional to perturbations in the angle of attack, α = Uw ), 0

q = angular rate about the positive y-axis, or pitch rate, θ = pitch-angle perturbation from the reference θo value, Xu,w,δe = partial derivative of the aerodynamic force in x direction with respect to perturbations in u, w, and δe,5 Zu,w,δe = partial derivative of the aerodynamic force in z direction with respect to perturbations in u, w, and δe, Mu,w,q,δe = partial derivative of the aerodynamic (pitching) moment with respect to perturbations in u, w, q, and δe, δe = movable tail-section, or “elevator,” angle for pitch control. Wo q, Uo q terms in the equations are due to the angular velocity of the body fixed (rotating) reference frame and arise directly from the left-hand side of Eq. (10.16). To determine altitude changes, we need to add the following equation to the longitudinal equations of motion: h˙ = Vref sin θ − w cos θ.

(10.23)

This equation will result in the linearized altitude equation h˙ = Vref θ − w, which is to be augmented with Eq. (10.22). 5 X, Z, M are stability derivatives and are identified from wind tunnel and flight tests.

(10.24)

10.3 Lateral and Longitudinal Control of a Boeing 747 For linearized lateral motion, the results are ⎤ ⎡ ⎤⎡ β˙ Yv −Uo Vo go cos θo ⎢ r˙ ⎥ ⎢ Nv ⎥⎢ Nr Np 0 ⎢ ⎥ ⎢ ⎥⎢ ⎣ p˙ ⎦ = ⎣ Lv ⎦⎣ Lr Lp 0 ˙ 0 0 tan θo 1 φ ⎡

where β = side-slip angle, defined to be

⎤ ⎡ β Yδr ⎢ Nδr r ⎥ ⎥+⎢ p ⎦ ⎣ Lδr 0 φ

689

⎤ Yδa   Nδa ⎥ ⎥ δr , Lδa ⎦ δa 0 (10.25)

v , Uo

r = yaw rate, p = roll rate, φ = roll angle, Yv,δr,δa = partial derivative of the aerodynamic force in the y direction with respect to perturbations in β, δr, and δa, Nv,r,p,δr,δa = aerodynamic (yawing) moment stability derivatives, Lv,r,p,δr,δa = aerodynamic (rolling) moment stability derivatives, δr = rudder deflection, δa = aileron deflection. We will next discuss the design of a stability-augmentation system for the lateral dynamics, called a yaw damper, and the autopilot affecting the longitudinal behavior.

10.3.1 Yaw Damper STEP 1. Understand the process and its performance specifications. Swept-wing aircraft have a natural tendency to be lightly damped in the lateral modes of motion. At typical commercial-aircraft cruising speeds and altitudes, this dynamic mode is sufficiently difficult to control that virtually every swept-wing aircraft has a feedback system to help the pilot. Therefore, the goal of our control system is to modify the natural dynamics so that the plane is acceptable for the pilot to fly.6 Studies have shown that pilots like natural frequencies ωn
0.5 and damping ratio of ζ ≥ 0.5. Aircraft with dynamics that violate these guidelines are generally considered fatiguing to fly and highly undesirable. Thus our system specifications are to achieve lateral dynamics that meet these constraints. STEP 2. Select sensors. The easiest measurement of aircraft motion to take is the angular rate. The side-slip angle can be measured with a wind-vane device, but it is noisier and less reliable for stabilization. Two angular rates—roll and yaw—partake in the lateral motion. Study of the lightly damped lateral mode indicates that it is primarily a yawing phenomenon, so measurement of the yaw rate is a logical starting point for the design. Until the early 1980s the measurement was made with a gyroscope with a small, fast-spinning rotor that can yield an electric output proportional to the angular yaw rate of the aircraft. Since the early 1980s most new aircraft systems 6 The mode is sufficiently difficult to control manually that, if the yaw damper fails in cruise, the pilot is

instructed to descend and slow down where the mode is more manageable.

690 Chapter 10 Control System Design: Principles and Case Studies have relied on a laser device (called a ring-laser gyroscope) for the measurement. Here, two laser beams traverse a closed path (often a triangle) in opposite directions. As the triangular device rotates, the detected frequencies of the two beams appear to shift, and this frequency shift is measured, producing a measure of rotational rate. These devices have fewer moving parts and are more reliable at less cost than the spinning-rotor variety of gyroscope. STEP 3. Select actuators. Two aerodynamic surfaces typically influence the lateral aircraft motion: the rudder and the ailerons (see Fig. 10.30). The lightly damped yaw mode that will be stabilized by the yaw damper is most affected by the rudder. Therefore, use of that single control input is a logical starting point for the design. Hence, it is best to choose the rudder as our actuator. Hydraulic devices are universally employed in large aircraft to provide the force that moves the aerodynamic surfaces. No other kind of device has been developed to provide the combination of high force, high speed, and light weight desirable for the actuation of the controlling aerodynamic surfaces. On the other hand, the low-speed flaps, which are extended slowly prior to landing, are typically actuated by an electric motor with a worm gear. For small aircraft with no autopilot, no actuator is required at all; the pilot yoke is directly connected to the aerodynamic surface by means of wire cables, and all the force required to move the surfaces is provided by the pilot. STEP 4. Make a linear model. The lateral-perturbation equations of motion for Boeing 747 in horizontal flight at 40,000 ft and nominal forward speed U0 = 774 ft/sec (Mach 0.8) (Heffley and Jewell, 1972), with the rudder chosen as the actuator (Step 3), are ⎡ ⎤ ⎡ ⎤⎡ ⎤ β˙ −0.0558 −0.9968 0.0802 0.0415 β ⎢ r˙ ⎥ ⎢ 0.598 ⎥⎢ r ⎥ −0.115 −0.0318 0 ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎣ p˙ ⎦ = ⎣ −3.05 ⎦⎣ p ⎦ 0.388 −0.4650 0 0 0.0805 1 0 φ φ˙ ⎡ ⎤ 0.00729 ⎢ −0.475 ⎥ ⎥ +⎢ ⎣ 0.153 ⎦ δr, 0 ⎡ ⎤ β ⎢ r ⎥ ⎥ y = [ 0 1 0 0 ]⎢ ⎣ p ⎦, φ where β and φ are in radians and r and p are in radians per second. The transfer function, using the MATLAB ss2tf function, is −0.475(s + 0.498)(s + 0.012 ± 0.488j) r(s) = , (10.26) δr(s) (s + 0.0073)(s + 0.563)(s + 0.033 ± 0.947j) so that the system has two stable real poles and a pair of stable complex poles. Notice first that the low-frequency gain is negative, corresponding to the simple physical fact that a positive or clockwise rudder motion causes a negative or counterclockwise yaw rate. In other words, turning the rudder left (clockwise) causes the front of the aircraft to rotate left (counterclockwise). The natural motion corresponding to the G(s) =

10.3 Lateral and Longitudinal Control of a Boeing 747 Dutch roll Spiral mode Roll mode

Washout

Washout

691

complex poles is referred to as the Dutch roll; the name comes from the motions of a person skating on the frozen canals of Holland. The motion corresponding to the stable real poles is referred to as the spiral mode (s1 = −0.0073) and the roll mode (s2 = −0.563). From looking at the system poles, we see that the offending mode that needs repair for good pilot handling is the Dutch roll, with the poles at s = −0.033 ± 0.95j. The roots have an acceptable frequency, but their damping ratio ζ ∼ = 0.03 is far short of the desired value ζ ∼ = 0.5. STEP 5. Try a lead–lag or PID design. As a first try at the design, we will consider proportional feedback of the yaw rate to the rudder. The root locus with respect to the gain of this feedback is shown in Fig. 10.32, and its frequency response is shown in Fig. 10.33. The figures show that ζ ∼ = 0.45 is achievable and can be computed to occur at a gain of about 3.0. This feedback, however, creates an objectionable situation during a steady turn when the yaw rate is constant: Because the feedback produces a steady rudder input opposite the yaw rate, the pilot must introduce a much larger steady command for the same yaw rate than is necessary in the open-loop case. This dilemma is solved by attenuating the feedback at DC (i.e., “washing out” the feedback). This is accomplished by inserting s H(s) = s + 1/τ in the feedback, which passes the yaw rate feedback at frequencies above 1/τ and provides no feedback at DC. Therefore, in a steady turn, the damper will provide no correction. Figure 10.34 shows a block diagram of the yaw damper with the washout. For a more complete model, we include the rudder servo, which represents the actuator dynamics and has the transfer function 10 δr(s) = , A(s) = eδr (s) s + 10 which is fast compared with the dynamics of the rest of the system and is not expected to change the response very much. The root locus, including actuator dynamics and a washout circuit with τ = 3, is shown in Fig. 10.35. As seen from the root locus, the addition of the yaw rate feedback, including the washout, allows the damping ratio to be increased from 0.03 to about 0.35. The associated frequency response of the system

Figure 10.32 Root locus for yaw damper with proportional feedback

Im(s) 2.0

2.0

1.0

1.0

2.0

2.0 Re(s)

692 Chapter 10 Control System Design: Principles and Case Studies 20 db

10

Magnitude, G

Figure 10.33 Bode plot of yaw damper with proportional feedback

0

1

20

0.1

0.01 0.1

40 10

1 v (rad/sec) (a)

100 50 Phase

0 50 100 150 180 0.1

1 v (rad/sec)

10

(b)

Figure 10.34 Yaw damper: (a) functional block diagram; (b) block diagram for analysis

da

Pilot’s aileron input to cause roll  Pilot’s rudder input to cause yaw rc



edr



Rudder servo

dr

Washout circuit

eyrg

e

Aircraft

r

Yaw rate gyro

(a) rc





 e

edr

10 s  10

dr

s s  1/t

eyrg

(b)

Aircraft

Kyrg

r

10.3 Lateral and Longitudinal Control of a Boeing 747 Figure 10.35 Root locus with washout circuit, τ = 3

693

Im(s) 1.0

10

5

0.5

1.0 Re(s)

8.47

1.0

is shown in Fig. 10.36. The response of the closed-loop system to an initial condition of β0 = 1◦ is shown in Fig. 10.37 for a root-locus gain of 2.6. For reference, the response of yaw rate with no feedback is also given. Although feedback of yaw rate through the washout circuit results in a considerable improvement over the original aircraft control, the response is not as good as originally specified. Further iterations, not included here, could include other gain values or more complex compensations. STEP 6. Evaluate/modify plant. The solution would be to unsweep the wings, which would cause a large drag penalty. STEP 7. Try an optimal design using pole placement. If we augment the dynamic model of the system by adding the actuator and washout, we obtain the state-variable model ⎤ ⎡ ⎤ ⎡ x˙ A −10 0 0 0 0 0 ⎥ ⎢ β˙ ⎥ ⎢ 0.0729 −0.0558 −0.997 0.0802 0.0415 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ r˙ ⎥ ⎢ −4.75 0.598 −0.1150 −0.0318 0 0 ⎥ ⎥=⎢ ⎢ ⎥ ⎢ p˙ ⎥ ⎢ 1.53 −3.05 0.388 −0.465 0 0 ⎥ ⎥ ⎢ ⎢ ⎦ ⎣ φ˙ ⎦ ⎣ 0 0 0.0805 1 0 0 0 0 1 0 0 −0.333 x˙ wo ⎡ ⎤ ⎤ ⎡ xA 10 ⎢ β ⎥ ⎢ 0 ⎥ ⎢ ⎥ ⎥ ⎢ ⎢ r ⎥ ⎢ 0 ⎥ ⎢ ⎥ ⎢ ⎥ eδr , ×⎢ ⎥+⎢ ⎥ ⎢ p ⎥ ⎢ 0 ⎥ ⎣ φ ⎦ ⎣ 0 ⎦ 0 x˙ wo ⎤ ⎡ xA ⎢ β ⎥ ⎥ ⎢ ⎢ r ⎥ ⎢ ⎥, e = [ 0 0 1 0 0 −0.333 ] ⎢ ⎥ ⎢ p ⎥ ⎣ φ ⎦ xwo where eδr is the input to the actuator and e is the output of the washout. The SRL for the augmented system is as shown in Fig. 10.38. If we select the state-feedback

694 Chapter 10 Control System Design: Principles and Case Studies 20

10

0

1

20

0.1

0.01 0.1

1

10

db

Magnitude, GHA

Figure 10.36 Bode plot of yaw damper, including washout and actuator

40

v (rad/sec) (a) 150 100

Phase

50 0 50 100 150 200 0.1

1

10

v (rad/sec) (b) 1 0.8

No feedback

0.6 0.4 b (deg)

Figure 10.37 Initial condition response with yaw damper and washout, and SRL design, for β0 = 1◦

0.2 0 0.2 0.4

SRL

0.6

Yaw rate feedback

0.8 1

0

5

10

15 20 Time (sec)

25

30

poles from the SRL so that the complex roots have maximum damping (ζ = 0.4), we find that pc = [−0.0051; −0.468; 0.279 + 0.628 ∗ j; 0.279 − 0.628 ∗ j; −1.106; −9.89] Then we can compute the state-feedback gain, using the MATLAB function place, to be K = [ 1.059 −0.191 −2.32 0.0992 0.0370 0.486 ].

10.3 Lateral and Longitudinal Control of a Boeing 747 Figure 10.38 SRL of lateral dynamics, including washout filter and actuator

695

Im(s) 2 24


0.4

10

1

1

10 Re(s)

Poles for maximum damping

Note that the third entry in K is larger than the others, so the feedback of all six state variables is essentially the same as proportional feedback of r. This is also evident from the similarity of the root locus in Fig. 10.31 and the SRL of Fig. 10.38. If we select the estimator poles to be five times faster than the controller poles, then pe = [−0.0253; −2.34; −1.39 + 3.14 ∗ j; −1.39 − 3.14 ∗ j; −5.53; −49.5] and the estimator gain, again using the MATLAB function place, is found to be ⎤ ⎡ 25.0 ⎢ −2,044 ⎥ ⎥ ⎢ ⎢ −5,158 ⎥ ⎥. ⎢ L=⎢ ⎥ ⎢ −24,843 ⎥ ⎣ −40,113 ⎦ −15,624 The compensator transfer function from Eq. (7.177) is Dc (s) =

Design trade-off: system response vs. system complexity

−844(s + 10.0)(s − 1.04)(s + 0.974 ± 0.559j)(s + 0.0230) . (10.27) (s + 0.0272)(s + 0.837 ± 0.671j)(s + 4.07 ± 10.1j)(s + 51.3)

Figure 10.37 also shows the response of the yaw rate to an initial condition of β0 = 1◦ . It is clear from the root locus that the damping can be improved by the SRL approach, and this is borne out by the reduced oscillatory behavior in the transient response of the system. However, this improvement has come at a considerable price. Note that the order of the compensator has increased from one in the original design (Fig. 10.33) to six and washout in the design obtained using the controller-estimator-SRL approach. Aircraft yaw dampers in use today generally employ a proportional feedback of yaw rate to rudder through a washout or through minor modifications to this design. The improved performance achievable with an optimal design approach utilizing fullstate feedback and estimation is not judged to be worth the increase in complexity.

696 Chapter 10 Control System Design: Principles and Case Studies Perhaps a more fruitful approach to improving the design would be to add the aileron surface as a control variable along with the rudder. STEP 8 and 9. Verify the design. Linear models of aircraft motion are reasonably accurate as long as the motion is small enough that the actuators and surfaces do not saturate. Because actuators are sized for safety in order to handle large transients, such saturation is very rare. Therefore, the linear-analysis-based design is reasonably accurate, and we will not pursue a nonlinear simulation or further design verification. However, aircraft manufacturers do carry out extensive nonlinear simulations and flight testing under all possible flight conditions before obtaining Federal Aviation Administration (FAA) certification to carry passengers.

10.3.2 Altitude-Hold Autopilot STEP 1. Understand the process and its performance specifications. One of the pilot’s many tasks is to hold a specific altitude. As an aid to keeping aircraft from colliding, those craft on an easterly path are required to be on an odd multiple of 1000 ft and those on a westerly path on an even multiple of 1000 ft. Therefore, the pilot needs to be able to hold the altitude to less than a hundred feet. A well-trained, attentive pilot can easily accomplish this task manually to within ±50 ft, and airtraffic controllers expect pilots to maintain this kind of tolerance. However, since this task requires the pilot to be fairly diligent, sophisticated aircraft often have an altitude-hold autopilot to lessen the pilot’s work. This system differs fundamentally from the yaw damper because its role is to replace the pilot for certain periods of time, while the yaw damper’s role is to help the pilot fly. Dynamic specifications, therefore, need not require that pilots like the craft’s “feel” (how it responds to their handling of the controls); instead, the design should provide the kind of ride that pilots and passengers like. The damping ratio should still be in the vicinity of ζ ∼ = 0.5, but for a smooth ride the natural frequency should be much slower than ωn = 1 rad/sec. STEP 2. Select sensors. Clearly needed is a device to measure altitude, a task most easily done by measuring the atmospheric pressure. Almost from the time of the first Wright brothers’ flight, this basic idea has been used in a device called a barometric altimeter. Before autopilots, the device consisted of a bellows whose free end was connected to a needle that directly indicated altitude on a dial. The same bellows concept is used today for the altitude display, but the pressure is sensed electrically for the autopilot. Because the transfer function from the controlling elevator input to the altitude control consists of five poles [see Eq. (10.30)], stabilization of the feedback loop cannot be accomplished by simple proportional feedback. Therefore the pitch rate q is also used as a stabilizing feedback; it is measured by a gyroscope or ring-laser gyro identical to that used for yaw-rate measurement. Further stabilization using pitchangle feedback is also helpful. It is obtained either from an inertial reference system based on a ring-laser gyro or from a rate-integrating gyro. The latter is a device similar to the rate gyro, but structured differently so that its outputs are proportional to the angles of the aircraft’s pitch θ and roll φ. STEP 3. Select actuators. The only aerodynamic surface typically used for pitch control on most aircraft is the elevator δe. It is located on the horizontal tail, well

10.3 Lateral and Longitudinal Control of a Boeing 747

697

removed from the aircraft’s center of gravity, so that its force produces an angular pitch rate and thus a pitch angle, which acts to change the lift from the wing. In some high-performance aircraft there are direct-lift control devices on the wing or perhaps small canard surfaces, which are like tiny wings forward of the main wing, which produce vertical forces on the aircraft that are much faster than elevators on the tail are able to generate. However, for purposes of our altitude hold, we will consider only the typical case of an elevator surface on the tail. As for the rudder, hydraulic actuators are the preferred devices to move the elevator surface, mainly because of their favorable force-to-weight ratio. STEP 4. Make a linear model. The longitudinal perturbation equations of motion for the Boeing 747 in horizontal flight at a nominal speed U0 = 830 ft/sec at 20,000 ft (Mach 0.8) with a weight of 637,000 lb are ⎡ ⎢ ⎢ ⎢ ⎢ ⎣

u˙ w˙ q˙ θ˙ h˙

x˙ = Fx + Gδe, ⎤ ⎡ −0.00643 0.0263 ⎥ ⎢ −0.0941 −0.624 ⎥ ⎢ ⎥ = ⎢ −0.000222 −0.00153 ⎥ ⎢ ⎦ ⎣ 0 0 0 −1 ⎡ ⎤ 0 ⎢ −32.7 ⎥ ⎢ ⎥ ⎥ +⎢ ⎢ −2.08 ⎥ δe, ⎣ ⎦ 0 0

−32.2 0 0 0 830

0 820 −0.668 1 0

0 0 0 0 0

⎤⎡

(10.28) ⎤

u ⎥⎢ w ⎥ ⎥ ⎥⎢ ⎥⎢ q ⎥ ⎥ ⎥⎢ ⎦⎣ θ ⎦ h

where the desired output for an altitude-hold autopilot is h = Hx,

h=[ 0

and

Phugoid mode Short-period modes

Inner-loop design

0

0

0

⎡ ⎢ ⎢ 1 ]⎢ ⎢ ⎣

u w q θ h

⎤ ⎥ ⎥ ⎥, ⎥ ⎦

32.7(s + 0.0045)(s + 5.645)(s − 5.61) h(s) = . δe(s) s(s + 0.003 ± 0.0098j)(s + 0.6463 ± 1.1211j)

(10.29)

(10.30)

The system has two pairs of stable complex poles and a pole at s = 0. The complex pair at −0.003 ± 0.0098j are referred to as the phugoid mode,7 and the poles at −0.6463 ± 1.1211 are the short-period modes, as computed using the MATLAB eig command. STEP 5. Try a lead–lag or PID controller. As a first step in the design, it is typically helpful to use an inner-loop feedback of pitch rate q to δe so as to improve the damping 7 The name was adopted by F. W. Lanchester (1908), who was the first to study the dynamic stability of

aircraft analytically. It is apparently an incorrect version of a Greek word.

698 Chapter 10 Control System Design: Principles and Case Studies of the short-period mode of the aircraft (see Fig. 10.39). The transfer function from δe to q, using the MATLAB ss2tf function, is 2.08s(s + 0.0105)(s + 0.596) q(s) =− . δe(s) (s + 0.003 ± 0.0098j)(s + 0.646 ± 1.21j)

(10.31)

The inner loop root locus for q feedback using Eq. (10.31) is as shown in Fig. 10.40. Because kq is the root-locus parameter, the system matrix [Eq. (10.28)] is now modified to (10.32) Fq = F + kq GHq , where F and G are defined in Eq. (10.28) and Hq = [ 0 0 1 0 0 ]. The process of picking a suitable gain kq is an iterative one. The selection procedure is the same one discussed in Chapter 5. (Recall the tachometer feedback example in Section 5.6.2.) If we choose kq = 1, then the closed-loop poles will be located at −0.0039 ± 0.0067j, −1.683 ± 0.277j on the root locus, and ⎤ ⎡ −0.00643 0.0263 0 −32.2 0 ⎢ −0.0941 −0.624 787.3 0 0 ⎥ ⎥ ⎢ ⎢ 0 0 ⎥ (10.33) Fq = ⎢ −0.000222 −0.00153 −2.75 ⎥. ⎣ 0 0 1 0 0 ⎦ 0 −1 0 830 0 Figure 10.39 Altitude-hold feedback system

Compensation href





Aircraft 

Dc(s)







de

h



q kq ku Inner loop

Figure 10.40 Inner-loop root locus for altitude-hold dynamics with q feedback

Im(s)

1

2

1

Re(s)

1 Poles for kq
1

u

10.3 Lateral and Longitudinal Control of a Boeing 747

699

Note that only the third column of Fq is different from F. To further improve the damping, it is useful to feed back the pitch angle of the aircraft. By trial and error, we select Kθ q = [ 0

0

−0.8

−6

0 ]

in order to feed back θ and q, and the system matrix becomes Fθq = Fq − GKθq , ⎡ −0.0064 0.0263 ⎢ −0.0941 −0.624 ⎢ =⎢ ⎢ −0.0002 −0.0015 ⎣ 0 0 0 −1

0 761 −4.41 1 0

−32.2 −196.2 −12.48 0 830

0 0 0 0 0

⎤ ⎥ ⎥ ⎥, ⎥ ⎦

with poles at s = 0, −2.25 ± 2.99j, −0.531, −0.0105. So far, the inner loop of the aircraft has been stabilized significantly. The uncontrolled aircraft has a natural tendency to return to equilibrium in level flight, as evidenced by the open-loop roots in the LHP. The inner-loop stabilization is necessary to enable an outer-loop feedback of h and h˙ to be successful; furthermore, the feedbacks of θ and q can be used by themselves in an attitude-hold mode of the autopilot, when a pilot wishes to control θ directly through input command. Figure 10.41 shows the response of the inner loop to a 2◦ (0.035-rad) step command in θ. With the inner loop in place the transfer function of the system from elevator angle to altitude is now h(s) 32.7(s + 0.0045)(s + 5.645)(s − 5.61) = . (10.34) δe(s) s(s + 2.25 ± 2.99j)(s + 0.0105)(s + 0.0531) The root locus for this system, given in Fig. 10.42, shows that proportional feedback of altitude by itself does not yield an acceptable design. For stabilization we may also feed back the rate of change in the altitude in a PD controller. The root locus of the system with feedback of both h and h˙ is shown in Fig. 10.43. After some iteration we find that the best ratio of h˙ to h is 10:1, that is, De (s) = Kh (s + 0.1). Figure 10.41 Response of altitude-hold autopilot to a command in θ

0.040 0.035 0.030

u(t) (rad)

0.025 0.020 0.015 0.010 0.005 0.000

0

1

2

3

4

5

6

7

8

9

10

700 Chapter 10 Control System Design: Principles and Case Studies Figure 10.42 0◦ root locus with feedback of h only

Im(s) 4

2

8

6

4

2

5.6

Re(s)

2 4

Figure 10.43 0◦ root locus with feedback of h and h˙

Im(s) 4 3 2 1 5 4 3 2 1

Re(s) 1 1 2 3 4

˙ and h feedback The final design is the result of iterations between the q, θ, h, gains, obviously a lengthy process. Although this trial design was successful, use of the SRL approach promises to expedite the process. STEP 6. Evaluate/modify plant. Not applicable here. STEP 7. Do an optimal design. The SRL of the system is shown in Fig. 10.44. If we choose the closed-loop poles at pc = [−0.0045; −0.145; −0.513; −2.25 − 2.98 ∗ j; −2.25 + 2.98 ∗ j] then the required feedback gain, using the MATLAB function place, is K = [ −0.0009

0.0016

−1.883

−7.603

−0.001 ].

The step response of the system to a 100-ft step command in h is shown in Fig. 10.45, and the associated control effort is shown in Fig. 10.46. This design has been carried out with the assumption that the linear model is valid for the altitude changes under consideration. We should perform simulations to verify this or to determine the range of validity of the linear model.

10.3 Lateral and Longitudinal Control of a Boeing 747 Figure 10.44 SRL for altitude-hold design

701

Im(s) 8 6 4 2

5 4 3 2 1

1

2

3

4

5

Re(s)

2 4 6 8

Figure 10.45 Step response of altitude-hold autopilot to a 100-ft step command

100

Altitude, h (ft)

80

60

40

20

0

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 Time (sec)

STEP 8. and 9. Verify the design. The comments in Steps 7 and 8 of Section 10.3.1 apply to this design as well. For small airplane autopilots now in production, such as the one described in Chapter 5, it is interesting to note that, for the inner loop, some manufacturers employ only θ feedback while others use q feedback. The use of θ enables faster response, but use of q is less costly. Both, of course, use the altimeter for h feedback.

702 Chapter 10 Control System Design: Principles and Case Studies Figure 10.46 Control effort for 100-ft step command in altitude

0 1 2 de 3 4 5 6

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 Time (sec)

10.4 Control of the Fuel–Air Ratio in an Automotive Engine Until the 1980s most automobile engines had a carburetor to meter the fuel so that the ratio of the gasoline-mass flow to air-mass flow, or fuel-to-air ratio (F/A), remained in the vicinity of 1:15. This device metered the fuel by relying on a pressure drop produced by the air flowing through a venturi. The device performed adequately in terms of keeping the engine running satisfactorily, but it historically allowed excursions of up to 20% in the F/A. After the implementation of federal exhaust-pollution regulations, this level of inaccuracy in the F/A was unacceptable because neither excess hydrocarbons (HCs) nor excess oxygen could be accepted. During the 1970s, automobile companies improved the design and manufacturing process of the carburetors so that they became more accurate and delivered a F/A accuracy in the vicinity of 3% to 5%.8 Through a combination of factors, this improved F/A accuracy helped lower the exhaust pollution levels. However, the carburetors were still open-loop devices because the system did not measure the F/A of the mixture entering the engine for subsequent feedback into the carburetor. During the 1980s almost all manufacturers turned to feedback control systems to provide a much-improved level of F/A accuracy, an action made necessary by the decreasing levels of allowable exhaust pollutants. We now turn to the design of a typical feedback system for engine control, again using the step-by-step design outline given in Section 10.1. STEP 1. Understand the process and its performance. The method chosen to meet the exhaust-pollution standards has been to use a catalytic converter that simultaneously oxidizes excess levels of exhaust carbon monoxide (CO) and unburned HCs and reduces excess levels of the oxides of nitrogen (NO and NO2 , or NOx ). This device is usually referred to as a three-way catalyst because of its effect on all three pollutants. This catalyst is ineffective when the F/A is much different from the stoichiometric 8A review of automotive engine control is contained in a book by Alexander Stotsky, Automotive Engine:

Control Estimation, Statistical Detection.

10.4 Control of the Fuel–Air Ratio in an Automotive Engine

703

level of 1:14.7; therefore, a feedback control system is required to maintain the F/A within ±1% of that desired level. The system is depicted in Fig. 10.47. The dynamic phenomena that affect the relationship between the sensed F/A output from the exhaust and the fuel-metering command in the intake manifold are (1) intake fuel and air mixing, (2) cycle delays due to the piston strokes in the engine, and (3) the time required for the exhaust to travel from the engine to the sensor. All these effects are strongly dependent on the speed and load of the engine. For example, engine speeds typically vary from 600 to 6000 rpm. The result of these variations is that the time delays in the system that will affect the feedback control-system behavior will also vary by at least 10:1, depending on the operating condition. The system undergoes transients as the driver demands more or less power through changes in the accelerator pedal, with the changes taking place over fractions of a second. Ideally, the feedback control system should be able to keep up with these transients. STEP 2. Select sensors. The discovery and development of the exhaust sensor was the key technological step that made possible this concept of exhaust-emission reduction by feedback control. The active element in the device, zirconium oxide, is placed in the exhaust stream, where it yields a voltage that is a monotonic function of the oxygen content of the exhaust gas. The F/A is uniquely related to the oxygen level. The voltage of the sensor is highly nonlinear with respect to F/A (Fig. 10.48); almost all the change in voltage occurs precisely at the F/A value at which the feedback system must operate for effective performance of the catalyst. Therefore, the gain of

Catalytic converter

Intake manifold Air

Engine

Exhaust Sensor

Actuator Fuel

Figure 10.47 F/A feedback control system Figure 10.48 Exhaust sensor output Sensor output, V

0.9

0.5

0.1 1:18

1:14.7 1:12 Fuel–air ratio, F/A

704 Chapter 10 Control System Design: Principles and Case Studies the sensor will be very high when the F/A is at the desired point (1:14.7) but will fall off considerably for F/A excursions away from 1:14.7. Although other sensors have been under development for possible use in F/A feedback control, no other cost-effective sensor has so far demonstrated the capability to perform adequately. All manufacturers of production-line automobiles currently use zirconium oxide sensors in their feedback control systems.

Nonlinear sensor

STEP 3. Select actuators. Fuel metering can be accomplished by a carburetor or by fuel injection. Implementing a feedback F/A system requires the capability of adjusting the fuel metering electrically, because the sensor used provides an electric output. Initially, carburetors were designed to provide this capability by including adjustable orifices that modify the primary fuel flow in response to the electric error signal. However, today manufacturers accomplish the metering by use of fuel injection. Fuel-injection systems are typically electrical by nature, so they can be used to perform the fuel adjustment for F/A feedback simply by including the capability of using the feedback signal from the sensor. Today, fuel injectors are placed at the inlet to every cylinder (called multipoint injection); in the past, there was one large injector upstream from all the cylinders (called single-point or throttle body injection). Multipoint injection offers improved performance because the fuel is introduced much closer to the engine, with better distribution to the cylinders. Being closer reduces the time delays and thus yields better engine response and enables lower exhaust pollution. STEP 4. Make a linear model. The sensor nonlinearity shown in Fig. 10.48 is severe enough that any design effort based on a linearized model of it should be used with caution. Figure 10.49 shows a block diagram of the system, with the sensor shown to have a gain Ks . The time constants τ1 and τ2 indicated for the inlet manifold dynamics represent, respectively, fast fuel flow in the form of vapor or droplets and slow fuel flow in the form of a liquid film on the manifold walls. The time delay is the sum of (1) the time it takes the pistons to move through the four strokes from the intake process until Inlet manifold Fast fuel Desired  F/A 

0.5

Control law



e

K Kp  I s

1 t1s  1

uf Slow fuel 0.5

Sensor y

Ks F/A

Figure 10.49 Block diagram of an F/A control system



esTd

 

1 t2s  1

Sensor lag 1 ts  1

Time delay Exhaust F/A

10.4 Control of the Fuel–Air Ratio in an Automotive Engine

705

the exhaust process and (2) the time required for the exhaust to travel from the engine to the sensor located roughly 1 ft away. A sensor lag with time constant τ is also included in the process to account for the mixing that occurs in the exhaust manifold. Although the time constants and the delay time change considerably, primarily as a function of engine load and speed, we will examine the design at a specific point where the values are τ1 = 0.02 sec, τ2 = 1 sec,

Complications of nonlinearity

Td = 0.2 sec, τ = 0.1 sec.

In an actual engine, designs would be carried out for all speed loads. STEP 5. Try a lead–lag or PID controller. Given the tight error specifications and the wide variations in the required fuel command uf due to varying engine-operating conditions, an integral control term is mandatory. With integral control, any required steady state uf can be provided when the error signal e = 0. The addition of a proportional term, although not often used, allows for an increase (doubling) in the bandwidth without degrading steady-state characteristics. In this example we use a control law that is proportional plus integral (PI). The output from the control law is a voltage that drives the injector’s pulse former to give a fuel pulse whose duration is proportional to the voltage. The controller transfer function can be written as Kp KI Dc (s) = Kp + = (s + z), (10.35) s s where KI z= Kp and z can be chosen as desired. First, let us assume that the sensor is linear and can be represented by a gain Ks . Then we can choose z for good stability and good response of the system. Figure 10.50 shows the frequency response of the system for Ks Kp = 1.0 and z = 0.3, while Fig. 10.51 shows a root locus of the system with respect to Ks Kp with z = 0.3. Both analyses show that the system becomes unstable for Ks Kp ∼ = 2.8. Figure 10.50 shows that to achieve a phase margin of approximately 60◦ , the gain Ks Kp should be ∼2.2. Figure 10.50 also shows that this produces a crossover frequency of 6.0 rad/sec (∼1 Hz). The root locus in Fig. 10.51 verifies that this candidate design will achieve acceptable damping (ζ ∼ = 0.5). Although this linear analysis shows that acceptable stability at a reasonable bandwidth (∼1 Hz) can be achieved with a PI controller, a look at the nonlinear sensor characteristics (Fig. 10.48) shows that this indeed may not be achievable. Note that the slope of the sensor output is extremely high near the desired setpoint, thus producing a very high value of Ks . Therefore, lower values of the controller gain Kp need to be used to maintain the overall Ks Kp value of 2.2 when including the effect of the high sensor gain. On the other hand, a value of Kp low enough to yield a stable system at F/A = 1 : 14.7 (= 0.068) will yield a very sluggish response to transient errors that deviate much from the setpoint, because the effective sensor gain will be reduced substantially. It is therefore necessary to account for the sensor nonlinearity in order to obtain satisfactory response characteristics of the system for anything other than minute disturbances about the setpoint. A first approximation to the sensor is shown

706 Chapter 10 Control System Design: Principles and Case Studies 20

Magnitude, DcG

10

0

1

2.2

2.8

20

0.1

0.01 0.1

0.4

1

2 4 10 20 40 v (rad/sec)

40 100

(a) 30 60

Phase

90 120 150

vc for PM
60

180 210 0.1

0.4

1

2 4 10 20 40 v (rad/sec)

100

(b)

Figure 10.51 Root locus of a PI F/A controller

Im(s) KsKp
2.8 
0.5

KsKp
2.2

50

20

15

10

5

5

KsKp
2.2 5

KsKp
2.8

Re(s)

db

Figure 10.50 Bode plot of a PI F/A controller

10.4 Control of the Fuel–Air Ratio in an Automotive Engine

707

in Fig. 10.52. Because the actual sensor gain at the setpoint is still quite different from its approximation, this approximation will yield erroneous conclusions regarding stability about the setpoint; however, it will be useful in a simulation to determine the response to initial conditions. STEP 6. Evaluate/modify plant. The nonlinear sensor is undesirable; however, no suitable linear sensor has been found. STEP 7. Try an optimal controller. The response of this system is dominated by the sensor nonlinearity, and any fine tuning of the control needs to account for that feature. Furthermore, the system dynamics are relatively simple, and it is unlikely that an optimal design approach will yield any improvement over the PI controller used. We will thus omit this step. STEP 8. Simulate design with nonlinearities. The nonlinear closed-loop simulation of the system implemented in SIMULINK
is shown in Fig. 10.53. The MATLAB function (fas) implements the approximate nonlinear sensor characteristics of Fig. 10.53, function y = fas(u) if u < 0.0606, Figure 10.52 Sensor approximation Sensor output, V

0.9

Setpoint Approximation

0.5

Actual sensor

0.1 1:18

1:14.7 1:12 Fuel–air ratio, F/A

Exhaust F/A

Scope1 Desired F/A   Step

0.5 e

Control law 0.1s  0.03 s

Gain

0.5 Gain1 Sensor nonlinearity MATLAB® Function

Fast fuel 1 0.02s  1 Slow fuel 1 s1 Sensor lag 1 0.1s  1

Figure 10.53 Closed-loop nonlinear simulation implemented in SIMULINK


 

Scope Time delay

708 Chapter 10 Control System Design: Principles and Case Studies y = 0.1 ; elseif u < 0.0741, y = 0.1 + (u − 0.0606) * 20; else y = 0.9; end Figure 10.54(a) is a plot of the system error using the approximate sensor of Fig. 10.52 and Kp Ks = 2.0. The slow response is apparent with 12.5 sec before the error comes out of saturation and a time constant of almost 5 sec once the linear region is reached. In real automobiles these systems are operated with much higher gains. To show these effects, a simulation with Kp Ks = 6.0 is plotted in Fig. 10.54(b, c). At this gain the linear system is unstable and up until about 5 sec the signals grow. The growth halts after 5 sec due to the fact that, as the input to the sensor nonlinearity gets large,

Figure 10.54 System response with nonlinear sensor approximation

0.4

e

0.2

0

5.0

10.0 Time (sec)

15.0

20.0

15.0

20.0

15.0

20.0

(a)

0.8 0.4 e 0 0.4

0

5.0

10.0 Time (sec) (b)

0.09 0.06 F/A 0.03 0

5.0

10.0 Time (sec) (c)

10.5 Control of the Read/Write Head Assembly of a Hard Disk

SIMULINK nonlinear simulation

709

the effective gain of the sensor decreases due to the saturation, and eventually, a limit cycle is reached. The frequency of this limit cycle corresponds to the point at which the root locus crosses the imaginary axis and has an amplitude such that the total effective gain of Kp Ks,eq = 2.8. As described in Section 9.3, the effective gain of a saturation for moderately large inputs can be computed and is given by the describing function to be approximately 4N/π a, where N is the saturation level and a is the amplitude of the input signal. Here N = 0.4, and if Kp = 0.1, then Ks,eq = 28. Thus we predict an input signal amplitude of a = 4(0.4)/28π = 0.018. This value is closely verified by the plot of Fig. 10.54(c), the input to the nonlinearity in this case. The frequency of oscillation is also nearly 10.1 rad/sec, as predicted by the root locus in Fig. 10.51. In the actual implementation of F/A feedback controllers in automobile engines, sensor degradation over thousands of miles of use is of primary concern, because the federal government mandates that the engines meet the exhaust-pollution standards for the first 50,000 mi. In order to reduce the sensitivity of the average setpoint to changes in the sensor output characteristics, manufacturers typically modify the design discussed here. One approach is to feed the sensor output into a relay function [see Fig. 9.6(b)], thus completely eliminating any dependency on the sensor gain at the setpoint. The frequency of the limit cycle is then solely determined by controller constants and engine characteristics. Average steady-state F/A accuracy is also improved. The oscillations in the F/A are acceptable because they are not noticeable to the car’s occupants. In fact, the F/A excursions are beneficial to the catalyst operation in reducing pollutants.

10.5 Control of the Read/Write Head Assembly of a Hard Disk The first mass storage device based on recording data on hard disks was introduced by IBM in 1956 as the model 350 RAMAC.9 It consisted of a stack of fifty 24-inch diameter aluminum disks that were coated with a magnetic material, and the data were recorded in concentric tracks at 100 bytes per inch with 20 tracks per inch. The disks were rotated at 1200 rpm. There was a single read/write head assembly mounted on an arm that could be moved vertically from disk to disk and horizontally across the chosen disk to reach a desired data track. The heads were held above the disk surface by an air bearing generated by blowing air through holes in the fixture holding the heads. The assembly was held on a particular disk by a detent on the elevator mechanism and held on a particular track by an arm detent. The entire head assembly was driven by a single electric motor. The system held 5 MB of data, and consideration had to be given to be sure that the final device could be passed through a door 36 in. wide. The technical advances in this field have been such that in the year 2000 Seagate introduced a hard drive magnetic memory consisting of three disks, each 2.5 inches in diameter, rotating at 15,000 rpm designed to be included in a portable laptop computer. This device could hold 18,350 megabytes of data. The read/write assembly consisted of a single arm moving a comb of heads, one per surface, in a rotary motion to move the heads from track to track. The heads are mountable on a gimbal at the end of the 9 Random Access Method of Accounting and Control

710 Chapter 10 Control System Design: Principles and Case Studies arm and fly above the surfaces of the disks. To follow a track, the assembly is under active feedback control using samples of position data recorded between the sectors of user data around each track. An economic measure of the progress in the field is that while the cost of the RAMAC data was about $10,000 per megabyte, that of a modern drive is less than 1 cent per megabyte. A brief summary of this remarkable history, with many references is given in Abramovitch and Franklin (2002), and a table of a few disk parameters over time is presented in Table 10.1. A large number of people from both industrial and academic institutions have contributed to the many technologies involved in the advances in hard-disk memory devices made over the past 50 years, and one of the enabling technologies has been feedback control. A picture of a Seagate 1000-GB disk drive is shown in Fig. 10.55. In this brief case study we will point out a number of issues involving control, but the design example will be concerned only with the issue of track following. We will follow the outline given in Section 10.1 in presenting the case. STEP 1. Understand the process. An exploded view of the track-following servo problem is given in Fig. 10.56. The mechanism consists of a rotary voice-coil motor moving an assembly of a light arm supporting gimbal-mounted sliders that include the magnetoresistive read heads and the light, thin-film inductive write heads. The slider flies above the disk surface on an air bearing produced by the disk rotation. The power amplifier is usually connected as a current amplifier so that the basic motion can be modeled as simple inertia, described by A (10.36) Go (s) = 2 , Js where J is the total inertia and A includes both the motor torque constant and the amplifier gain. The structure is flexible, however, and the detailed motion is very complex, with many lightly damped modes. It is also subject to buffeting from the air flow and from vibration caused by housing motion. For purposes of control design, a single resonant mode will be included according to the model   s +1 2ζ A ω1 , G(s) = 2  (10.37) 2 Js s s + 2ζ +1 ω1 ω1 where the vibration frequency, ω1 , and the damping ratio, ζ , are known only within bounds. The motion control of the head assembly is in two modes: the seek motion to move the head from track to track and the track-follow motion to maintain the heads over the center of the selected track. In the seek mode the criterion is minimum time, and theory would call for “on–off”or “bang-bang”10 control. In order to use the same controller for many units, which differ in the maximum torque available and other critical parameters, the method used in disk drives is a bang-curve-follow technique in which the assembly is accelerated under full torque until the velocity reaches a torque reversal curve based on the distance to the desired track and deceleration 10 Common names for the case in which the control is saturated with one polarity for half the time, then

reversed for the remaining half.

TABLE 10.1

Disk Drive Parameters Over Time

No.

Year

Unit

Capacity

Size (N/d)

tpi

bpi

rpm

1 2 3

1956 1962 1971

IBM RAMAC IBM 1301 IBM 3330

5 MB 28 MB 100 MB

50/24" 25/24" 11/14"

20 50 192

100 520 4040

1200 1800

4

1973

70 MB

4/14"

270

5600

5

1979

3340 Winchester IBM 3370

571 MB

7/14"

635

12,134

6

1979

IBM 3310

64.5 MB

6/8"

450

8530

7 8

1980 1983

SeagateST506 MaxtorXT1140

5 MB 126 MB

4/5.25" 8/5.25"

255

7690

9 10 11 12

1991 1993 1997 2000

1 GB 2.19 GB 4 GB 18.3 GB

8/3.5" 10/3.5" 3/2.5" 5/2.5"

2238

58,874

12,500 21.5k

211,000 343k

15,000

13

2003

300 GB

4/3.3"

105k

658k

10,000

14

2006

300 GB

4/2.75"

125k

890k

15,000

15

2007

IBM Corsair Seagate 12550 IBMTravelstar Seagate ST318451 Seagate ST3300007 Seagate ST3300655 Barracuda ES.2

1000 GB

4/3.75"

150k

1090k

7,200

Fly Height 20 µ 1.2 µ 0.5 µ

2964

0.324 µ

Head Type

Sensor Type

Actuator Type

Air bearing Flying head Ferrite,

Detent Detent Dedicated surface Dedicated surface Dedicated surface Hybrid, sector servo Open loop Sector servo

dc motor Hydraulic piston Linear voice coil flying Linear voice coil

Sector servo Sector servo Sector servo Sector servo

Rotary voice coil Rotary voice coil Rotary voice coil Rotary voice coil

Sector servo

Micro-actuator

Sector servo

Rotary voice coil

Sector servo

Rotary voice coil

Ferrite, flying Thin film

MR head 7200

GMR = giant magnetoresitive head; TMR = tunneling magnetoresistive head.

Thin-film/ GMR Thin-film/ GMR Thin-film/ GMR Thin-film/ TMR

Seek Time

Comment The first hard disk

165 ms 30 ms

The first feedback Low-mass heads

Linear voice coil Rotary voice coil

27 ms

Stepper motor Rotary voice coil

170 ms

5.25" disk for PCs In-hub spindle motor

3.9(r), 4.5(w) 4.9(r), 5.4(w) 3.5(r), 4.0(w) 7.4(r), 8.5(w)

First 15,000 RPM disk drive First microactuators First perpendicular recording drive First SAS drive

711

712 Chapter 10 Control System Design: Principles and Case Studies Figure 10.55 Picture of 1000-GB disk drive Source: Courtesy Seagate Technology LLC

Position sensing noise Magnetic media

PES

DAC noise

ADC noise

Spindle dynamics 



 

DAC

ADC DSP

Disk position

Motor torque

Windage (air flow)







Head position

Figure 10.56 Generalized view of track-following model

Arm mechanics

10.5 Control of the Read/Write Head Assembly of a Hard Disk

713

is under feedback control to follow this curve to reach the desired track with zero velocity. The curve approximates the optimal minimum time switching curve with torque discounted to the extent that the weakest motor will have a reserve of torque adequate to follow the curve. When the selected track is reached, the control transfers to track-following mode. A scheme to avoid mode switching when the selected track is approached and to cause the servo to move seamlessly into track-follow mode has been called the Proximate Time Optimal Servo or PTOS (see Chapter 9).11 As a mature technology, many trends have influenced the nature of the control problem over the years. For example, as the table shows, disks have become smaller and thus stiffer and smoother. As the arm assembly has become smaller, it has less inertia to the extent that for very small motions as in a one- or two-track transfer, friction is more important than inertia. For recent drives, the width of a track is on the order of 0.2 micron (µ), a value comparable to the feature dimensions on a modern integrated circuit chip! To counter this trend, research is exploring ways to add a second actuator, either on the arm or on the gimbal, to make small moves much as the wrist acts on the end of a robot arm. Because of the difficulty of controlling a very lightly damped flexibility, consideration is also given to adding a coating to the arm to increase the damping of the principal modes of vibration. Other proposals include adding sensors on the arm to allow extra feedback to control the flexibility. In this case study, we will assume a single voice-coil actuator and that the flexibility is described as in Eq. (10.37), where ω1 ≥ 2π × 2.500 and ζ ≥ 0.05. Because the details of the actual resonance are not well known, the resonance will need to be gain stabilized. STEP 2. Select sensors. The earliest drives were controlled open loop with one mechanical detent to hold the assembly on a disk and another detent to hold the heads on a track. Feedback control was introduced in 1971 using position information recorded on a special disk surface dedicated to the servo data. The entire comb of heads was positioned by the servo surface information. If the comb were to tilt or otherwise be misaligned, the data would be that much more difficult to read. Such issues limited the number of disks and the track density possible with this arrangement. The track position information in modern disks is recorded on each track in a gap between the sectors of user data. Controls based on this information are called sector servos, and the data are sampled of necessity. There is a conflict between the desire to record large amounts of data, which calls for fewer and larger sectors, and the control requirement to have a high sample rate, which calls for smaller sectors. Each case is a compromise between these conflicting demands. Because the position data are sampled, the controllers are digital devices to make the best possible use of the position data. Theoretical study has been given to using a multirate control to apply more than one control correction for each sensor reading, but the method has not been found to be cost effective yet. For the case study here, we will design an analog controller. The position information extracted from data recorded on the disk is subject to errors caused by run-out in the track path, which means that the radius of the track is not constant. In general, there is a repeatable component in each trip around the track, and this element can be estimated, often harmonic by harmonic, and a signal used as 11 Workman (1987), Franklin, Powell, and Workman (1998).

714 Chapter 10 Control System Design: Principles and Case Studies feed-forward to the motor to cancel it out. The position error signal (PES) also contains random noise from many sources. These include the buffeting by the airflow over the slider, wobble and vibration of the disks, noise in the signal-processing electronics used to decode the position information, noise from the power amplifier used to provide torque to the motor, and errors caused by the analog-to-digital converters needed in the process. STEP 3. Select actuators. The RAMAC used a DC motor as actuator, and later drives used hydraulic actuators. When the 5.25-in. drive was introduced by Seagate in 1980, the actuator was a stepping motor. Each of these were used in open loop. The first feedback control of the head position was on the IBM 3330 in 1971, and the actuator was a linear-motion voice-coil motor. In 1979 a rotary voice-coil motor was introduced, and today almost all hard disk drives use a rotary motion actuator. The power amplifier is usually connected as a current amplifier to simplify the dynamics. The feedback from the current-sensing resistor to the amplifier constitutes a “torque loop” that is designed separately and carefully, so that the dynamics of the motor can be ignored most of the time in considering the outer loop position control in track following. STEP 4. Make a linear model. As mentioned in the discussion of the process, the linear model has one flexible mode, namely G(s) =

1 s

2



(2ζ s/ω1 + 1)

,

s2 s + 2ζ +1 ω1 ω1

(10.38)

where we take ζ = 0.05 and ω1 = 2.5, corresponding to measuring time in milliseconds rather than seconds. The gain A and the inertia J will be absorbed in the gain of the compensator. The power amplifier is thus assumed to be an ideal current amplifier. Also we are considering only track following, and not seek. STEP 5. Try a PID or lead–lag design. Because the nominal model is so simple, the first design will be a lead compensation with the objective of achieving the greatest possible bandwidth subject to having a phase margin of 50◦ and such that it will gain stabilize the resonance with a gain margin of at least 4. This approach was already published by R. K. Oswald (1974). We will try two designs and compare them for bandwidth and the quality of their step responses. In the first case, we will use a simple lead compensation, selected to give 50◦ phase margin and a factor of 4 gain margin. To get the phase margin, the lead will be designed with an α of 0.1, and the crossover frequency will be placed as high as possible while keeping a gain margin of 4 at the resonance, which rises by a factor of 1/2ζ = 10 above the Bode asymptote. Thus the crossover must be located so that the asymptote is a factor of 10 × 4 = 40 below 1 at ω1 = 5π . The resulting lead transfer function is D(s) = 0.617

(2.22s + 1) , (0.222s + 1)

(10.39)

and the Bode plot of the lead design is shown in Fig. 10.57. The gain crossover frequency for this design is ωc = 1.39 rad/msec and the step response is plotted in Fig. 10.58, which shows a rise time of about tr = 0.8 msec with an overshoot of about 25%. We have shown before that a phase margin of 50◦ should correspond to a damping of 0.5 and thus an overshoot of about 17%. However,

Figure 10.57 The Bode plot of the design with a single lead

Magnitude (db)

10.5 Control of the Read/Write Head Assembly of a Hard Disk 80 60 40 20 0 20 40 60 101

715

GM 100

101

102

101

102

v (rad/sec)

Phase (deg)

100 140

PM

180 220 260 101

100 v (rad/sec)

Figure 10.58 Step response of disk drive control with PM = 50◦

1.4 1.2

Amplitude

1 0.8 0.6 0.4 0.2 0

0

0.8

1.6

2.4 3.2 4 Time (msec)

4.8

5.6

because the zero of the lead is in the forward path, we get the extra overshoot that goes with such a zero. As a second design, a roll-off filter is to be added to try to suppress the resonance peak in order to gain a bit in speed of response and bandwidth. The idea is to put the filter cutoff frequency between the crossover frequency and the resonance frequency and to give it a damping ratio low enough that it does not reduce the phase margin too much but high enough that it does not interfere with the gain margin. After some experimentation, the trial design D(s) = 1.44

(1.48s + 1) (0.148s + 1)

(10.40)

is tested with a filter of F(s) =

1 s2

s +1 + 0.6 2 (10.3) 10.3

.

(10.41)

For this case the Bode plot is given in Fig. 10.59 and the step response in Fig. 10.60.

716 Chapter 10 Control System Design: Principles and Case Studies In this case the crossover frequency is 2.13, a 35% increase, and the rise time is 0.3 msec, a 60% reduction from the case without the roll-off filter. The overshoot is a bit higher in this case. Although not presented here, further possibilities for the control compensation might include a notch filter rather than the low pass filter designed here. A notch might be able to further suppress the resonance and permit further increase in the bandwidth. A great deal depends on the degree of understanding of the resonance and how much uncertainty surrounds its behavior. In some cases, it is possible to phase-stabilize the resonance and to raise the crossover to be higher than the resonance frequency.

Figure 10.59 Bode plot of system with lead plus roll-off filter

Magnitude (db)

STEP 6. Evaluate/modify plant. Possible changes to the process that involve major design changes were introduced in the discussion concerning understanding the process in Step 1 above. Once the major parameters of the design have been selected, the remaining possibilities for improvement might include a change in the fabrication of the arm to add stiffness, which will raise the frequency of the vibration, and to add a damping coating to the arm to increase the damping ratio of the flexibility. Other 80 40 GM

0 40 101

100

101

102

101

102

v (rad/sec)

Phase (deg)

100 140

PM

180 220 260 101

100 v (rad/sec)

Figure 10.60 Step response of system with lead plus roll-off filter

1.4 1.2

Amplitude

1 0.8 0.6 0.4 0.2 0

0

0.8

1.6

2.4 3.2 4 Time (msec)

4.8

5.6

10.6 Control of RTP Systems in Semiconductor Wafer Manufacturing Figure 10.61 Step response for LQR design

717

1.4 1.2

Amplitude

1 0.8 0.6 0.4 0.2 0

0

0.2

0.4

0.6 0.8 Time (msec)

1

1.2

1.4

possibilities for improvement concern changes in the PES decoding methodology to reduce the noise. STEP 7. Try an optimal controller or adaptive control. A design was done with the linear quadratic performance measure with the performance index (loss function) selected to obtain a rise time of about 0.3 msec to match the classical design. The result is shown in Fig. 10.61. Although further effort might produce an acceptable design, the clearly oscillatory response tolerated by this particular technique does not look promising. In particular, a design that includes a cost on y˙ as well as y should be considered. Such extensions are considered in more advanced courses. STEP 8. Simulate the design, and compare the alternatives. Usually done in parallel with the design. STEP 9. Build a prototype. Done early in the design process as a bench model so trial schemes can be tested on hardware as designed. For digital control design and implementation of disk drive servos, the reader is referred to Franklin, Powell, and Workman (1998).

10.6 Control of RTP Systems in Semiconductor Wafer Manufacturing Figure 10.62 diagrams the major steps in the manufacture of an ultra-large-scale integrated circuit such as a microprocessor and some of the associated control aspects. Many of the steps described in this process, such as chemical vapor deposition or etching, must be done at closely controlled and timed temperature sequences (Sze, 1988). The standard practice for many years has been to perform these steps in batches on many wafers at a time to produce large numbers of identical chips. In response to the demand for ever smaller critical dimensions of the devices on the chip, and to give more flexibility in the variety and number of chips to be produced, the makers of the tools for fabrication of integrated circuits are asked to provide more and more precise control of temperature and time profiles during thermal processing. In response to these demands, an important trend is to perform the thermal steps on one wafer at a

718 Chapter 10 Control System Design: Principles and Case Studies

Figure 10.62 Steps in making an integrated circuit Source: Courtesy International Sematech

10.6 Control of RTP Systems in Semiconductor Wafer Manufacturing RTP

719

time in a chamber with cold walls and a flexible heat source called a rapid thermal processor (RTP) as shown in Fig. 10.63. The demands on an RTP system are illustrated by the requirement that it cause the wafer temperature to follow a profile such as that shown in Fig. 10.64, where the ramp-up speeds are at rates of 25◦ to 150◦ C/sec, and the soak temperatures range from 600◦ C to 1100◦ C and last from a few to as many as 120 sec. The ramp-up rates

Figure 10.63 Applied Materials’ Radiance RTP system Source: Courtesy Applied Materials

Figure 10.64 Typical RTP temperature trajectory Temperature

High-temperature soak Low-temperature soak

Warm-up

Time

720 Chapter 10 Control System Design: Principles and Case Studies

Pyrometer

Tungsten halogen lamp

Figure 10.65 Generic RTP system

are limited by the danger of causing damage to the crystal structure if the temperature gradients become too large. The ability of the RTP to change temperature rapidly permits fabrication of devices with very small critical lengths by being able to stop the processes such as deposition or etching quickly and accurately. Figure 10.65 shows a generic RTP reactor with tungsten halogen lamps, stainless steel walls that are water cooled, and quartz windows. Temperature measurement can be done by a variety of methods, including thermocouples, RTDs, and pyrometers. For various reasons (particle generation, minimal disturbance, etc.), it is desirable to use noncontact temperature sensing; therefore, pyrometric techniques are the most commonly employed. A pyrometer is a noncontact temperature sensor that measures infrared (IR) radiation, which is directly a function of the temperature. It is known that objects emit radiant energy proportional to T 4 , where T is the temperature of the object. Among the advantages of pyrometers are that they have very fast response time, and can be used to measure the temperature of moving objects (e.g., a rotating semiconductor wafer), and in vacuum for semiconductor manufacturing. The selection of the actuator depends on the choice of techniques for supplying power (tungsten halogen lamps, arc lamps, hot susceptor, etc.) to heat the wafer. Tungsten halogen lamps are now commonly used in RTP in semiconductor manufacturing (Emami-Naeini et al., 2003). Figure 10.66(a) shows a system with two-sided heating by linear tungsten halogen lamps (typical of systems produced by Mattson). The lamp arrays on the top and bottom are at right angles to provide more of an axisymmetry. Fig. 10.66(b) shows one-sided heating with lamps in a honeycomb configuration (typical of the Applied Materials systems). Finally Fig. 10.66(c) shows a configuration of lamps arranged in concentric rings (typical of the Stanford–TI MMST chamber, Gyugyi et al., 1993). The lamps do saturate and, for practical reasons, it is desired to operate them within 5%–95% of power settings. To illustrate the design of an RTP system, we give the results of a specific design carried out at SC Solutions as a laboratory model constructed to study problems associated with RTP design and operation. The laboratory model is shown schematically in Fig. 10.67. It is made of aluminum. It consists of three standard 35-W 12-V tungsten halogen lamps heating a rectangular plate that simulates the wafer. The plate

Centerline ytop ylamp

r1

r2

r3

Top wall Lamps

r4 dlamp

r5

rwall Top side wall

dywin

Quartz window ywin ysh

Showerhead rsh

ywaf

dysh

Wafer

Bottom side wall

dywaf yexit

Exit

Bottom wall rexit

10.6 Control of RTP Systems in Semiconductor Wafer Manufacturing Figure 10.66 Various lamp geometries for RTP

721

Wafer Lamps

Window edges

Lamps

Lamps

Source: Norman, 1992

(a)

Figure 10.67 Block diagram of the RTP laboratory model

(b)

(c)

Tungsten halogen lamps

PMW amplifier

Sensor electronics

Temperature sensors

Feedback controller

Te





Vcmd

Ty 

Noise filter

Tr

Reference generator

measures 4 in. × 1 43 in. and is blackened to increase its radiation absorption. The plate is mounted parallel to the lamps. The lamps are mounted in the lamp housing. The lamp assembly is mounted on a railing so that the distance from the lamps to the plate is adjustable. As the lamps are moved out the gain of the system decreases, but the radiation cross talk (coupling) increases. On the other hand, as the lamps are moved closer to the plate, the gain of the system increases and the coupling is reduced. The nominal distance from the lamps to the plate is 1 in., but it is adjustable to several inches. The lamps are driven by a pulse-width modulated (PWM) amplifier driver. There is a separate power supply unit. There are three dials mounted on the

722 Chapter 10 Control System Design: Principles and Case Studies

RTP laboratory model

side for open-loop and manual system operations. There are 14 resistive temperature detector (RTD) strips mounted vertically behind the back of the plate: 12 on the plate and 2 on each support on either side. There is a noise source filter that generates periodic sensor noise at 1.5 Hertz so as to represent noise seen in real RTP systems. All electronics (i.e., sensor signal processing and PWM amplifier) reside in the enclosure at the bottom of the unit. Because there is exposure to the outside, the surrounding environment provides sources of disturbance. STEP 1. Understand the process and its performance specifications. RTP is an inherently dynamic and nonlinear process. Among interesting properties of the system are multiple time scales (time constants for lamps, wafer, showerhead, and quartz window are different); nonlinear (radiation dominant) behavior; nonlinear lamps; effects of power supplies; number and placement of sensors; number, placement, and grouping of lamps; and large temperature variations. The DC gain in the system (δ temperature/δ power) decreases with increasing temperature due to the nonlinear increase in radiative losses. Various types of physical models are needed. Detailed physical models are required for equipment design, but reduced-order models are needed for fast evaluation of geometry changes, recipe development, and for feedback control design. Smooth transition between manual and automatic control is also required. STEP 2. Select sensors. This was discussed earlier. For the laboratory model, the sensors were a set of 14 RTDs, but three (located at the center and the support edges of the plate) can be used for feedback and the rest can be used for temperature monitoring purposes. In our case, we will use only the center temperature for feedback control. (Another alternative would be to sum the three temperatures into one signal and control the average temperature.) STEP 3. Select actuators. This was also discussed earlier. For the laboratory model, the actuators were composed of three standard tungsten halogen lamps previously described. In our case, we shall tie up all three lamps into a single actuator by applying the same input command to each lamp.

Nonlinear radiation heat transfer

STEP 4. Make a linear model. The laboratory model was built (see Step 9). The nonlinear system equations involve both conduction (Chapter 2) and radiation terms (see Emami-Naeini et al., 2003). Nonlinear system identification approaches were used to derive a model for the system. Specifically, the three lamps were stepped up, held constant, and then stepped down sequentially, and the three output temperatures were recorded. System identification studies12 resulted in the following nonlinear model for the system that contains the radiation and conduction terms (Ar and Acon respectively):  4   T T ˙ + B u. (10.42) M T = Ar + Acon T∞ T∞ Here T = [T1 T2 T3 ]T denote the temperatures, T∞ = constant ambient temperature (T˙ ∞ = 0), u = [vcmd1 vcmd2 vcmd3 ]T are the voltage commands, and the system matrices are 12 Performed by Dr. G. van der Linden.

10.6 Control of RTP Systems in Semiconductor Wafer Manufacturing ⎤ 1.000040 0 0 ⎦, 0 5.557443 0 M−1 = ⎣ 0 0 13.638218 ⎡ 5.4762e − 2 −8.5706e − 3 −8.2961e − 4 −4.5361e − 2 8.5709e − 3 −1.6213e − 7 −8.9134e − 8 Ar = ⎣ −8.5706e − 3 −8.2961e − 4 −1.6213e − 7 8.2998e − 4 2.0976e − 7 ⎡ 3.5599e − 7 −1.1136e − 7 −1.1976e − 7 −4.7011e − 8 1.1602e − 2 −2.5027e − 3 −9.0992e − 3 Acon = ⎣ −1.1136e − 7 −1.9761e − 7 −2.5027e − 3 6.3736e − 3 −3.8707e − 3 ⎡ ⎤ 3.4600e − 1 1.1772e − 1 2.8380e − 2 B = ⎣ 3.8803e − 11 8.0249e − 2 1.8072e − 2 ⎦ . 8.0041e − 9 2.7216e − 3 3.1713e − 2

723

⎡

⎤ ⎦, ⎤ ⎦,

A linear model for the system was derived as T˙ = F3 T + G3 u,

(10.43)

y = H3 T + J3 u, RTP linear model

where y = [Ty1 Ty2 Ty3 and ⎡ ⎤ −0.0682 0.0149 0.0000 0.0218 ⎦, F3 = ⎣ 0.0458 −0.1181 0.0000 0.04683 −0.1008 ⎡ ⎤ 1 0 0 H3 = ⎣ 0 1 0 ⎦, 0 0 1 ]T

⎡

0.3787 0.1105 G3 = ⎣ 0.0000 0.4490 0.0000 0.0007 ⎡ ⎤ 0 0 0 J3 = ⎣ 0 0 0 ⎦. 0 0 0

⎤ 0.0229 0.0735 ⎦, 0.4177

The three open-loop poles are computed from MATLAB and are located at −0.0527, −0.0863, and −0.1482. For our case, because we tied the three lamps into one actuator and are using only the center temperature for feedback, the linear model is then ⎤ ⎡ ⎤ ⎡ −0.0682 0.0149 0.0000 0.5122 F = ⎣ 0.0458 −0.1181 0.0218 ⎦ , G = ⎣ 0.5226 ⎦ , 0.4185 0.0000 0.04683 −0.1008  H= 0 1 0 , J = [0] , resulting in the transfer function G(s) =

Ty2 (s) 0.5226(s + 0.0876)(s + 0.1438) = . Vcmd (s) (s + 0.1482)(s + 0.0527)(s + 0.0863)

STEP 5. Try a lead–lag or PID controller. We may try a simple PI controller of the form (s + 0.0527) , Dc (s) = s so as to cancel the effect of one of the slower poles. The linear closed-loop response is shown in Fig. 10.68(a) and the associated control effort is shown in Fig. 10.68(b). The system response follows the commanded trajectory with a time delay of

30

5

25

0

20

r

Lamp voltage

Temperature (K)

724 Chapter 10 Control System Design: Principles and Case Studies

y

15 10

5 10 15 20

5 0

u

0

10 20 30 40 50 60 70 80 90 100 Time (sec)

25

0

10 20 30 40 50 60 70 80 90 100 Time (sec)

(a) Temperature tracking response

(b) Control effort

Figure 10.68 Linear closed-loop RTP response for PI controller approximately 2 sec and no overshoot. The lamp has its normal response until 75 sec and goes negative (shown dashed) to try to follow the sharp drop in commanded temperature. This behavior is not possible in the system, as there is no means of active cooling and the lamps do saturate low. Note that there is no explicit means of controlling the temperature nonuniformity here. STEP 6. Evaluate/modify plant. This was discussed already in connection with actuator and sensor selection. STEP 7. Try an optimal design. We use the error-space approach for inclusion of integral control and employ the linear quadratic Gaussian technique of Chapter 7. The error system is        e˙ 0 H e J = + µ, (10.44) 0 F ξ G ξ˙ 

where A=

0 0

H F



 ,

B=

J G

 ,

e = y − r, ξ = T˙ , and µ = u˙ . For state feedback design, the LQR formulation of Chapter 7 is used; that is,

∞ J = {zT Q z+ρµ2 } dt, 0

ξ T ]T .

Temperature uniformity

where z = [e Note that J needs to be chosen in such a way as to penalize the tracking error e and the control u, as well as the differences in the three temperatures. Therefore, the performance index should include a term of the form 2 1 10 (T1 − T2 )2 + (T1 − T3 )2 + (T2 − T3 )2 , which minimizes the temperature nonuniformity. The factor of 10 was determined by trial and error as the relative weighting between the error state and the plant state.

10.6 Control of RTP Systems in Semiconductor Wafer Manufacturing

725

The state and control weighting matrices, Q and R, respectively, are then ⎡ ⎤ 1 0 0 0 ⎢ 0 20 −10 −10 ⎥ ⎥ Q=⎢ ⎣ 0 −10 20 −10 ⎦ , R = ρ = 1. 0 −10 −10 20 The following MATLAB command is used to design the feedback gain: [K] = lqr(A,B,Q,R). The resulting feedback gain matrix computed from MATLAB is K = [K1 : K0 ], where K1 = 1,

K0 =



0.1221

2.0788

−0.2140



,

which results in the internal model controller of the form x˙ c = Bc e,

(10.45)

u = Cc xc − K0 T, with xc denoting the controller state and Bc = −K1 = −1, Cc = 1. The resulting state-feedback closed-loop poles computed from MATLAB are at −0.5574 ± 0.4584j, −0.1442, and −0.0877. The full-order estimator was designed with the process and sensor noise intensities selected as the estimator design knobs: Rw = 1,

Rv = 0.001.

The following MATLAB command is used to design the estimator: [L] = lqe(F,G,H,Rw,Rv). The resulting estimator gain matrix is ⎡

⎤ 16.1461 L = ⎣ 16.4710 ⎦ , 13.2001

with estimator error poles at −16.5268, −0.1438, and −0.0876. The estimator equation is ˙ˆ ˆ + Gu+L(y−HT). ˆ T = FT (10.46) With the estimator, the internal model controller equation is modified as x˙ c = Bc e,

(10.47)

ˆ u = Cc xc − K0 T. The closed-loop system equations are given by x˙ cl = Acl xcl + Bcl r, y = Ccl xcl + Dcl r,

(10.48)

726 Chapter 10 Control System Design: Principles and Case Studies where r is the reference input temperature trajectory, the closed-loop state vector is ˆ T ]T and the system matrices are xcl = [TT xcT T ⎡ ⎡ ⎤ ⎤ F GCc 0 −GK0 ⎦ , Bcl = ⎣−Bc ⎦ , 0 0 Acl = ⎣Bc H LH GCc F − GK0 −LH 0  Ccl = H 0 0 , Dcl = [0], with closed-loop poles (computed with MATLAB) located at −0.5574 ± 0.4584j, −0.1442, −0.0877, −16.5268, −0.1438 and −0.0876 as expected. The closed-loop control structure is shown in Fig. 10.69. The closed-loop control system diagram implemented in SIMULINK is shown in Fig. 10.70. The linear closed-loop response is shown in Fig. 10.71(a), and the associated control effort is shown in Fig. 10.71(b). The commanded temperature trajectory, r, is a ramp from 0◦ C to 25◦ C, with a 1◦ C/sec slope followed by 50-sec soak time and a drop back to 0◦ C. (Note that the ramp rate is very slow here because we have only three lamps for our RTP laboratory model, whereas a real RTP system would Linear plant  R

1 s

Bc





xc

Cc



u





x
Fx  Gu y
Hx

Y

Estimator Dc

xˆ
(F  LH) xˆ  Gu  Ly

Figure 10.69 Closed-loop control structure diagram

Scope 1

Scope 2 k1

  Repeating sequence

k1 • u

RTP plant

Integrator 1 s

x
Ax  Bu y
Cx  Du

 

Matrix gain1

Saturation

State-space

k0

Estimator

k0 • u Matrix gain

Figure 10.70 SIMULINK
block diagram for RTP closed-loop control

x
Ax  Bu y
Cx  Du State-space1

Scope

10.6 Control of RTP Systems in Semiconductor Wafer Manufacturing 30

20 r

y

Lamp voltage

Temperature (K)

25

15 10 5 0 5

0

10 20 30 40 50 60 70 80 90 100 Time (sec) (a) Temperature tracking response

6 4 2 0 2 4 6 8 10 12 14

727

u

0

10 20 30 40 50 60 70 80 90 100 Time (sec) (b) Control effort

Figure 10.71 Linear closed-loop RTP response

Temperature trajectory following

SIMULINK nonlinear simulation Prefilter

Lamp nonlinearity

have hundreds of lamps, and the much faster ramp rates mentioned earlier would be relevant.) The system tracks the commanded temperature trajectory—albeit with a time delay of approximately 2 sec for the ramp and a maximum of 0.089◦ C overshoot. As expected, the system tracks a constant input asymptotically, with zero steady-state error. The lamp command increases as expected to allow for tracking the ramp input, reaches a maximum value at 25 sec, and then drops to a steady-state value around 35 sec. The normal response of the lamp is seen from 0 to 75 sec, followed by a negative commanded voltage for a few seconds corresponding to fast cooling. Again, the negative control effort voltage (shown in dashed lines) is physically impossible as there is no active cooling in the system. Hence, in the nonlinear simulations, commanded lamp power must be constrained to be strictly nonnegative (Step 8). Note that the response from 75 to 100 sec is that of the (negative) step response of the system. STEP 8. Simulate the design with nonlinearities. The nonlinear closed-loop system was simulated in SIMULINK as shown in Fig. 10.72a. The model was implemented in temperature units of degrees Kelvin and the ambient temperature is 301K.13 The nonlinear plant model is the implementation of Eq. (10.42). There is a prefilter following the reference temperature trajectory (to smoothen the sharp corners) with the transfer function 0.2 Gpf (s) = . (10.49) s + 0.2 Note that conversion from voltage to power was determined experimentally to be given by P = V 1.6 (10.50) and is implemented as a nonlinear block (named VtoPower) in the SIMULINK diagram accordingly. The inverse of the static nonlinear lamp model is also included as a block 13 [K] = [◦ C] + 273.

728 Chapter 10 Control System Design: Principles and Case Studies To workspace 1 r

To workspace 2 u

Scope 2

Scope 3

Scope 1

k1  

InvLamp  

K•u

Repeating sequence

Integrator 1 s

 

In 1 Saturation

Matrix Gain 1

Out 1

Subsystem 2

VtoPower In 1

Out 1

Subsystem 1

Antiwindup K•u

 

Matrix Gain 2 k0

Scope 4

Estimator x
Ax  Bu y
Cx  Du State-space 1

K•u Matrix Gain

K•u Matrix Gain 4

To workspace 2 y

Lamp uncertainty Nonlinear plant Scope

Out 1

K•u

In 1

Subsystem

Add flux In 1

K•u Matrix Gain 4

Matrix Gain 5

Matrix Gain 3

(a)

Input flux 1

K•u

M_inv   Sum 

K•u

Integrator 1 xo
301.35 s

K•u Matrix Gain

Radiation

Node temp

In 2 Scale temp

Boltzmann Math function uy

1

K•u

Mux 3

Matrix Gain 2 Constant 1 4

Matrix Gain 5 K•u Matrix Gain 3 (b)

Mux 2

Constant 301.352135

Conduction

Figure 10.72a SIMULINK
diagram for nonlinear closed-loop RTP system: (a) nonlinear closed-loop; (b) nonlinear plant

10.6 Control of RTP Systems in Semiconductor Wafer Manufacturing Figure 10.72b SIMULINK
diagram for nonlinear closed-loop RTP system: (c) subsystem to convert voltage to power; (d) subsystem for lamp model inversion

 

1 In 1

Math function uy

Saturation

729

1 Out 1

Constant 1

Constant 1 1.6 (c)

1

Math function uy

In 1

 

1 Out 1

Constant 1 0.625

Constant 1 (d)

4

335

3.5

325

r

Lamp voltage (V)

Temperature (K)

330

y

320 315

3

u

2.5 2.0 1.5 1 0.5

310 20

40

60

80 100 120 140 160 180 Time (sec)

(a) Temperature tracking response

0 20

40

60

80 100 120 140 160 180 Time (sec) (b) Control effort

Figure 10.73 Nonlinear closed-loop response

(named InvLamp): V = P0.625 .

(10.51)

This will cancel the lamp nonlinearity. The voltage range for system operation is between 1 and 4 volts, as seen from the diagram. A saturation nonlinearity is included for the lamp as well as integrator antiwindup logic to deal with lamp saturation. The nonlinear dynamic response is shown in Fig. 10.73(a) and the control effort is shown in Fig. 10.73(b). Note that the nonlinear response is in general agreement with the linear response.

730 Chapter 10 Control System Design: Principles and Case Studies Figure 10.74 RTP temperature control laboratory model Source: Photo courtesy of Abbas Emami-Naeini

A prototype of the RTP laboratory model was designed, built,14 and was demonstrated at the Sematech AEC/APC’98 Conference, in Vail, Colorado. Figure 10.74 shows a photograph of the operational system. This system is really multivariable in nature. The three-input–three-output multivariable controller used on the prototype system was designed using the same approach discussed in Step 7, and was implemented on an embedded controller platform that uses a real-time operating system. The continuous controller (i.e., the combined internal model controller and the estimator) is of the form x˙ c = Ac xc + Bc e, (10.52) where

xc

=

[xcT

ˆ T ]T , T Ac =

u = Cc xc , 

0 GCc

 0 , F − GK0 −LH

 Bc =

 Bc , L

(10.53)

and

 Cc = Cc −K0 . The controller was discretized (see Chapter 8) with a sampling period of Ts = 0.1 sec and implemented digitally (with appropriate antiwindup logic) as c = c xkc +
c ek , (10.54) xk+1 uk = Cc xkc . The response of the actual system to the reference temperature trajectory, along with the three lamp voltages, is shown in Fig. 10.75. It is in good agreement with the nonlinear closed-loop simulation of the system (once noise is accounted for). For further information on modeling and control of RTP systems, the reader is referred to Emami-Naeini et al. (2003), Ebert et al. (1995a,b), de Roover et al. (1998), and Gyugyi et al. (1993). 14 By Dr. J. L. Ebert.

10.7 Chemotaxis or How E. Coli Swims Away from Trouble 65 60 Temperatlure (C)

Figure 10.75 Response of the RTP temperature control laboratory model

731

Tr 55 50 Ty2 45 Ty1 40 Ty3

35 0

20

40

60 Time (sec)

80

100

120

140

120

140

3.5 Vcmd1

Vcmd (V)

3 Vcmd3 2.5 2 Vcmd2 1.5 1

0

20

40

60 Time (sec)

80

100

10.7 Chemotaxis or How E. Coli Swims Away from Trouble

Systems Biology

E.coli

Background The cell is the basic structural and physiological subsystem of all living organisms and most of the biochemical activities necessary for life are performed in cells. Some organisms, such as bacteria, consist of only a single cell. A prokaryotic cell is shown in Fig. 10.76. Escherichia coli (E. coli), photographed in Fig. 10.77, is one of these single-cell organisms that has been extensively studied and whose interesting motion and control will be described in a highly simplified way in this case study. The technical results for the study come from the field of systems biology. Systems biology is an emerging field with the goal of creating dynamic models to describe the incredibly complex processes in many biological systems. The aim is to determine how shifting variables in one part impacts the whole. In this case study, a highly simplified model is presented to suggest how ideas from control can contribute to this effort. In preparing the study, we have tried to minimize the use of technical terms from biology and to define clearly those found useful and necessary for the presentation. It is hoped that this simple introduction will inspire control engineers to conduct direct study of this important field. First, a bit of background. E. coli was discovered by German pediatrician and bacteriologist Theodor Escherich in 1885. The bacterium is a cylindrical organism with hemispherical endcaps similar to the sketch shown in Fig. 10.76. A photograph of E. coli is shown

732 Chapter 10 Control System Design: Principles and Case Studies

Figure 10.76 A cell structure: (a) a typical bacterium; (b) TEM of bacterium Bacillus coagulans Source: (a) Campbell and Reece, page 98, 2008. © Pearson Education; (b) © Stanley C. Holt/Biological Photo Service. All rights reserved

Figure 10.77 Photograph of Escherichia coli (E. coli) bacteria Source: United States Department of Health and Human Services, National Institutes of Health

in Fig. 10.77. It is approximately 1 micron (µ) in diameter and 2 microns (µ) in length and weighs about 1 picogram (pg). E. coli has been studied extensively by geneticists because of its rather small genome size and the ease of growth in the laboratory. The entire genome, or the “library” of inherited genetic information, has been sequenced: it contains 4,639,221 of the adenosine (A), cytosine (C), guanine (G), and thymine (T) nitrogenous bases arranged into a total of 4288 genes. These genes serve as instructions for the synthesis of specific proteins, and are transcribed and eventually translated into the primary structure, or amino acid sequence, of a protein. E. coli grows longer and divides by binary fission to create two genetically identical “daughter” bacteria. It is a “cell division machine” and divides continuously such that under optimal conditions, a population of E. coli can double every 20

10.7 Chemotaxis or How E. Coli Swims Away from Trouble

Chemotaxis

733

minutes. In 2003 researchers demonstrated that solitary E. coli cells exhibit positive chemotaxis, which means that they are attracted to like cells enabling formation of E. coli colonies. E. coli lives in the lower intestine of warm blooded animals including humans and feed on amino acids. The bacterium helps in maintaining the balance of normal intestinal flora (bacteria) against harmful bacteria and synthesizing or producing some vitamins. Most E. coli strains are harmless but a particular strain (E. coli O157:H7) can cause food poisoning in humans. Escherischia coli has a set of 6 to 10 rotary motors each driving a thin helical filament about 10 µm long through a short, flexible and proximal hook that acts as a universal joint. This entire assembly is called a flagellum (Berg, 2003). The motor runs either clockwise (CW), as seen by an observer outside the cell looking down at the hook, or counterclockwise (CCW). When all the motors rotate CCW, the flagella filaments bundle together and the cell swims steadily forward in a “run” as suggested in Fig. 10.78. When one or more motors switches to CW rotation, the corresponding flagella unbundle and reorient the cell in a “tumble” resulting in little displacement as shown in Fig. 10.79. The two modes of motion alternate and in a state of equilibrium with its environment the runs last about 1 sec and the tumbles about 0.1 sec resulting in a 3-D random walk. Through control of tumbling frequency, the bacteria can direct their motion toward a concentration of attractant molecules or away from a concentration of repellent molecules as suggested in Fig. 10.80. The Problem Chemotaxis is the name given to the process by which a motile bacterium tastes the changes in its environment and moves toward places with a more favorable environment. Chemotaxis is important for proper functioning of the cell. An E. coli bacterium compares the current attractant concentration with the past attractant concentration. If it detects a positive change in the attractant concentration, it should move up the gradient. To do so, the probability of a tumble and hence its tumbling frequency is reduced and the runs are correspondingly longer. In contrast, if it detects an increase in repellent concentration, the assumption seems to be that it must have been swimming in a bad direction so it increases its tumbling frequency and tries to change

Figure 10.78 Flagella motors turning CCW resulting in a run

Plasma membrane

Source: Courtesy Nima Cyrus Emami

Cell wall

E. coli Flagella “Bundle”

734 Chapter 10 Control System Design: Principles and Case Studies Figure 10.79 Flagella motors turning CW resulting in a tumble Source: Courtesy Nima Cyrus Emami

Cell wall

Plasma membrane

Figure 10.80 Escherschia coli movements resembling a biased random walk

Ligand

direction so as to swim away from the repellents. The dynamics of this chemotaxis is the subject of our case study. Several different models of bacterial chemotaxis have been developed by researchers in systems biology. Our discussion is based on two of these (Barkai & Libler, 1997; Yi et al, 2000). The different proteins involved in chemotactic response have been well studied and their interactions have been characterized in some detail as shown in Fig. 10.81. Biologists have named the proteins involved in Chemotaxis by letters of the alphabet prefixed by “Che.” Thus we have CheA, CheB, and so on to CheZ. On the surface of the bacterium are receptor complexes, which include the CheW and CheA, to which the attractant or repellent molecules may bind. These chemicals constitute the input to the system and are called collectively ligands. The system is set up to control the frequency of tumbling, which is done by control of the activity of CheY, the protein that acts directly on the motor of the flagella.

10.7 Chemotaxis or How E. Coli Swims Away from Trouble Figure 10.81 The chemotaxis signal transconduction pathway in E. coli

735

Input (attractants, repellents)

CheR

m p

CheB

CheW

CheW

CheA

CheA

CheB

p

CheY

CheY

CheZ Output (tumbling frequency)

The receptors are either active and awaiting a ligand or are inactive and not accepting any ligand. A receptor complex becomes active if a methyl group (–CH3) is added to it by CheR and inactive if the group is removed by CheB. The CheR level is mainly fixed while the CheB level is controlled by the activity of the receptor via CheA. As part of the steady-state dynamics of chemotaxis, methyl groups are regularly being added by CheR and equally removed by CheB. This balance is upset when a ligand binds to an active receptor. If the ligand is an attractant, the activity of CheA is reduced and consequently the action of CheB in demethylation is reduced, more receptors are made active and the activity of CheA slowly returns to the steady state. This is the feedback loop in chemotaxis. At the same time as it reduces the rate of activation of CheB, CheA reduces its rate of activating CheY and this causes the tumbling frequency to be reduced. As a consequence, the bacteria swim more and presumably swim toward the attractant concentration. Now, if the ligand is a repellent, the activity of CheA is increased, which causes increased rate of CheY activity and increased frequency of tumbling. The bacteria swims less while it “looks” for a new direction in order to escape the concentration of repellents. At the same time, in the feedback loop, CheB is also more active, receptors are made inactive at a greater rate and again CheA and the tumble frequency return to their steady-state values. The fact that the activity and the tumble frequency return to exactly the same value after a change in ligand concentration is a remarkable property called by system biologists exact adaptation. As we will see, to a control engineer, this is a very common control method. An experimental plot of chemotaxis is reproduced in Fig. 10.82.

736 Chapter 10 Control System Design: Principles and Case Studies Figure 10.82 Experimental data of E. coli chemotaxis (Berg, 1972). The plots are planar projections of 3-D paths

AW405 Wild type 29.5s 26 runs Mean speed 21.2 µm/s

50 µm

The Model The problem, then, is to develop a model as a control system block diagram that will describe the average motion of this chemotaxis situation. We represent the averages as if they were one receptor complex with the related proteins acting on the flagella. As the research shows, the equations are complex and highly nonlinear. Also, the surface of the bacterium contains hundreds of receptor complexes and these interact as suggested already in Fig. 10.81. For our study, the variables for the block diagram are selected as linear, small signal deviations of the averages of the several quantities away from their equilibrium values. The input is taken to be the concentration of ligand, with attractors being positive and repellents being negative. The outputs of the system are the activity of CheA–P and resulting motion in the single x direction. The parameters of our model were selected so the responses matched the curves given in Fig. 10 of (Mello et al. 2004). The mechanics of one-dimensional motion assume that the viscous friction dominates the mass so the dynamics are a single integrator. The model is based on the following facts. •

•

• Adaptation

It is observed that when a ligand binds to an active receptor site, the changes in concentrations of CheA–P and resulting CheB and CheY–P are almost instantaneous. However, the changed CheB concentration only changes the rate of demethylation, not the extent of de-methyl itself. The changes in methyl level take place much more slowly than the changes in tumble rate. Upon insertion of a concentration of attractants, the “activity” as measured by the concentration of CheA drops quickly, then slowly recovers to exactly the same steady-state level. This property is called adaptation of activity.

10.7 Chemotaxis or How E. Coli Swims Away from Trouble Figure 10.83 Simplified block diagram of E. coli chemotaxis.  represents ligand, m the methylation, CheR the steady-state rate of methylation, and w the steady-state random walk motion

Y l





ka

CheA CheB 

 m

ky 

 CheY−P  

737

W kx



1 s



x

CheR

km s

Activity  

0.5

Ligand

K



1 s Integrator

2.0 Km



0.5

-K-

Ka

0.5 Kx

1.0

.5

CheR

Constant

Random number



1 s Motion

x

Scope

KM 

1.1 Mo

Methyl

Figure 10.84 A SIMULINK schematic diagram for simulating E. coli chemotaxis

A control block diagram shown in Fig. 10.83 implements these facts, including the adaptation. As seen, the adaptation result is accomplished by the standard control scheme of integral control. A SIMULINK schematic is shown in Fig. 10.84 and the responses in Figures 10.85, 10.86 and 10.87 for fixed concentrations of CheR. If the value of CheR is changed, the steady-state intensity of the activity changes and the time constant of the methylation also changes. In the end, we leave this case study with more questions than answers. For example, one should be able to derive the model by a small signal analysis from the basic chemical and physical equations of the processes. The model as presented could be modified to account for changes in the concentration of CheR, for example. Finally, how would the model be extended to describe the motion in three dimensions? We hope someone using this book is inspired to find the answers to these questions. Summary and Recap For years biologists had been focusing on studying various parts of living organisms. Recently, the focus has shifted to studying the whole organism’s behavior as a system

738 Chapter 10 Control System Design: Principles and Case Studies 1 0.9 0.8 0.7 Tumble Activity

Figure 10.85 Simulated tumble frequency of the chemotaxis model following insertion of attractant at t = 20 sec

Attractant added

0.6 0.5 0.4 0.3 0.2 0.1 0

Figure 10.86 Methylation of the chemotaxis model following insertion of attractant at t = 20 sec

0

5

10

15

20 25 Time (sec)

30

35

40

0

5

10

15

20 25 Time (sec)

30

35

40

2.9 2.8

Methylation level

2.7 2.6 2.5 2.4 2.3 2.2 2.1 2

of interconnected parts. Since the 1970s, it had been known experimentally that many biological systems adjust to their environment in an adaptive way. Recently, analytical models have been developed to explain this phenomenon as we discussed in this case study. The new analytical models can explain the inherent properties of the biological system such as robust perfect adaptation as given by the integral control of the active sites. Control theory methods and interpretations have proved helpful in increasing the level of our understanding of the behavior and properties of biological systems. We hope that this simple example helps stimulate interest in this exciting field.

10.8 Historical Perspective Figure 10.87 Motion response of the chemotaxis model following insertion of attractant at t = 20 sec

739

0.45 0.4 0.35

Position x

0.3 0.25 0.2 0.15 0.1 0.05 0

0

5

10

15

20 25 Time (Sec)

30

35

40

10.8 Historical Perspective The first autopilot was tested on a Curtis flying boat in 1912, just 9 years after the first Wright brothers’ flight. It consisted of a gyroscope to measure the attitude and servo motors to activate the control surfaces and was designed by Elmer Sperry. In part, it was a result of the Wright brothers design to intentionally make the aircraft slightly unstable in order to make it more controllable by the pilot. This system gained fame in 1914 when it won a prize in France by demonstrating its capabilities by flying close to the ground with the mechanic walking back and forth along the wing with the pilot, Lawrence Sperry, standing in the cockpit with his hands in the air.15 Autopilot development went underground in 1915 due to military security for WWI. The next public display was an adaptation of the Sperry system for Wiley Post in his 1933 flight around the world in “Winnie Mae.” The flight would have been near impossible without the autopilot because it allowed Post to doze off on occasion. It has been reported that Post had a system consisting of a wrench and string tied to his finger that would wake him up if he slept too soundly. The success of this flight led to the development of an autopilot that included some navigational capabilities as well as attitude control and, in 1947, the Air Force demonstrated an automatic trans-Atlantic flight in a DC-3 type airplane from take-off to landing. Subsequently, airplanes developed swept wings and higher speeds which required stability augmentation systems to help the pilot control the aircraft even when not on the autopilot. These systems are on all high-performance military and commercial airliners today. In 1974, the F-16 became the first airplane to have aerodynamically unstable regimes and was, therefore, highly dependent on the stability augmentation for sustained flight. This was implemented in order to make the airplane more

15 No mechanical connection from the stick to the control surfaces.

740 Chapter 10 Control System Design: Principles and Case Studies maneuverable, but required a “fly-by-wire” and quad redundancy for acceptable reliability. The first spacecraft in the late 1950s had no attitude control since their only mission was to take measurements and broadcast the information back to earth. However, they were followed in the early 1960s with the Corona spacecraft whose mission was to take photographs of the earth, which required that the camera be pointed and stabilized very accurately. At the time, these missions were classified for military purposes and called Discoverer for public consumption, but since then have been declassified and described in some detail.16 The first digital autopilots were in the Apollo program lunar module and the command module in the late 1960s. They were developed primarily by MIT’s Instrumentation Lab under the direction of Bill Widnall, Don Fraser, and Dick Battin. The decision to take the bold step of using digital technology for the first time rather than the traditional analog implementation was made by NASA in order to handle the complexity required at a reasonable weight. Prior to 1980, automobile engine control systems consisted of a mechanical arrangement in the distributor to vary the spark timing and a fluidic system in the carburetor that varied fuel flow in response to the airflow rate or sudden changes in the accelerator pedal position. These were open-loop systems that essentially programmed the proper control setting based on the operating condition of the engine. In 1980, cars were required to improve their polluting characteristics; therefore, it was essential to improve the controls by using feedback as described in Section 10.4. These systems still exist today along with variable valve timing, variable fuel injection timing, and variable valve opening levels. Application of control to semiconductor wafer manufacturing automation is gaining momentum. Many important process steps such as RTP, chemical–mechanical planarization, and lithography use advanced real-time controllers. It is anticipated that during the next decade many more of the semiconductor fabrication equipment will employ sophisticated in-situ feedback control as new sensors become available. This adoption of sophisticated closed-loop control systems by the semiconductor industry presents new challenges and opportunities for control system engineers especially for the upcoming 450-mm diameter wafers. Application of control to magnetic resonance force microscopy (MRFM) for imaging atomic structure of materials (de Roover et al., 2008) can fundamentally change our understanding of atomic structures of devices and enable imaging of biological subsystems. The emerging field of systems biology marks the coming of age of the life sciences. The usual approach of studying individual components is being replaced by a new approach focused on understanding the behavior of the whole biological system. Among the admirable goals are understanding the behavior of biological systems and discovering cure for diseases such as cancer, as well as developing novel approaches to discovery of new drugs, production of antibiotics, and vaccines. The applications of control theory have never been more exciting than they are today. Applications of feedback control ideas to biological systems, network congestion control, and new aerospace systems are emerging. Applications in genomics

16 Taubman (2003).

Summary

741

treating the human body as a dynamic system are underway. The Internet has attracted the attention of many control systems researchers eager to understand the tremendous success of this technology and how to improve it. Network design and control including Internet modeling, and development of congestion routing and control are under study. A number of our colleagues are enthusiastic about application of control theory to the financial field. If you like that, you will love the real-estate bubble!

SUMMARY •

In this chapter we have laid out a basic outline of control systems design and applied it to six typical case studies. The design outline calls for a number of explicit steps. 1. Make a system model and determine the required performance specifications. The purpose of this step is to answer the question, What is the system, and what is it supposed to do? 2. Select sensors. A basic rule of control is that if you can’t observe it, you can’t control it. Following are some factors to consider in the selection of sensors: (a) Number and location of sensors; (b) Technology to be used; (c) Performance of the sensor, such as its accuracy; (d) Physical size and weight; (e) Quality of the sensor, such as lifetime and robustness to environment changes; (f) Cost. 3. Select actuators. The actuators must be capable of driving the system so as to meet the required performance specifications. The selection is governed by the same factors that apply to sensor selection. 4. Make a linear model. All our design methods are based on linear models. Both small-signal perturbation models and feedback-linearization methods can be used. 5. Try a simple PID controller. An effort to meet the specifications with a PID or its cousin, the lead–lag compensator, may succeed; in any case such an effort will expose the nature of the control problem. 6. Evaluate/modify plant. Evaluate whether plant modifications enhance closed-loop performance; if so, return to Step 1 or 4. 7. Try an optimal design. The SRL method for control-law selection and estimator design based on state equations is guaranteed to produce a stable control system and can be structured to show a trade-off between error reduction and control effort. A related alternative is arbitrary pole placement, which gives the designer direct control over the dynamic response. Both the SRL and the pole-placement methods may result in designs that are not robust to parameter changes. 8. Simulate the design, and verify its performance. All the tools of analysis should be used here, including the root locus, the frequency response, GM

742 Chapter 10 Control System Design: Principles and Case Studies and PM measurements, and transient responses. Also, the performance of the design can be tested in simulation against changes in model parameters and the effects of approximating the compensator with a discrete model if digital control is to be used. 9. Build a prototype, and measure the performance with typical input signals. The proof of the pudding is in the eating, and no control design is acceptable until it has been tested. No model can include all the features of a real physical device; so the final step before fixing the design is to try it out on a physical prototype if time and budget permit. •

•

• •

• • • •

The satellite case study illustrated particularly the use of a notch compensation for a system with lightly damped resonance. It was also shown that collocated actuator and sensor systems are much easier to control than the noncollocated systems. The Boeing 747 lateral-stabilization case study illustrated the use of feedback as an inner-loop designed to aid the pilot, who provides the primary outer-loop control. The Boeing 747 altitude control showed how to combine inner-loop feedback with outer-loop compensation to design a complete control system. The automobile fuel–air ratio control illustrated the use of the Bode plot to design a system that includes time delay. Simulation of the design with the nonlinear sensor verified our heuristic analysis of limit cycles using the concept of equivalent gain with a root locus. The disk-drive case study illustrated control in an uncertain environment, where bandwidth is very important. The RTP case study illustrated modeling and control of a nonlinear thermal system. The E. coli chemotaxis case study illustrated a simple example of the application of ideas from control theory to the emerging field of systems biology. In all cases the designer needs to be able to use multiple tools, including the root locus, the frequency response, pole placement by state feedback, and simulation of time responses to get a good design. We promised an understanding of these tools at the beginning of the text, and we trust you are now ready to practice the art of control engineering.

REVIEW QUESTIONS 1. Why is a collocated actuator and sensor arrangement for a lightly damped structure such as a robot arm easier to design than a noncollocated setup? 2. Why should the control engineer be involved in the design of the process to be controlled? 3.

Give examples of an actuator and a sensor for the following control problems: (a) Attitude control of a geosynchronous communication satellite. (b) Pitch control of a Boeing 747 airliner. (c) Track-following control of a CD player. (d)

Fuel–air ratio control of a spark-ignited automobile engine.

Problems (e)

Position control for an arm of a robot used to paint automobiles.

(f)

Heading control of a ship.

743

(g) Attitude control of a helicopter.

PROBLEMS 10.1

Of the three types of PID control (proportional, integral, or derivative), which one is the most effective in reducing the error resulting from a constant disturbance? Explain.

10.2

Is there a greater chance of instability when the sensor in a feedback control system for a mechanical structure is not collocated with the actuator? Explain.

10.3

Consider the plant G(s) = 1/s3 . Determine whether it is possible to stabilize this plant by adding the lead compensator Dc (s) = K

s+a , s+b

(a < b).

(a) What is the maximum phase margin of the resulting feedback system? (b) Can a system with this plant, together with any number of lead compensators, be made unconditionally stable? Explain why or why not. 10.4

Consider the closed-loop system shown in Fig. 10.88. What is the phase margin if K = 70,000? What is the gain margin if K = 70,000? What value of K will yield a phase margin of ∼70◦ ? What value of K will yield a phase margin of ∼0◦ ? Sketch the root locus with respect to K for the system, and determine what value of K causes the system to be on the verge of instability. (f) If the disturbance w is a constant and K = 10,000, what is the maximum allowable value for w if y(∞) is to remain less than 0.1? (Assume r = 0.) (g) Suppose the specifications require you to allow larger values of w than the value you obtained in part (f) but with the same error constraint [|y(∞)| < 0.1]. Discuss what steps you could take to alleviate the problem.

(a) (b) (c) (d) (e)

Figure 10.88 Control system for Problem 10.4

d(s) R(s)

 

10.5



K(s  1) s  100







1 s(s  5)(s  10)

Y(s)

Consider the system shown in Fig. 10.89, which represents the attitude rate control for a certain aircraft. (a) Design a compensator so that the dominant poles are at −2 ± 2j. (b) Sketch the Bode plot for your √ design, and select the compensation so that the crossover frequency is at least 2 2 rad/sec and PM ≥ 50◦ . (c) Sketch the root locus for your design, and find the velocity constant when ωn > √ 2 2 and ζ ≥ 0.5.

744 Chapter 10 Control System Design: Principles and Case Studies Figure 10.89 Block diagram for aircraft-attitude rate control

Compensator R





Hydraulic servo

Aircraft

1 s

2s  0.1 s2  0.1s  4

Dc(s)



Y

Rate gyro krg

10.6

Consider the block diagram for the servomechanism drawn in Fig. 10.90. Which of the following claims are true? (a) The actuator dynamics (the pole at 1000 rad/sec) must be included in an analysis to evaluate a usable maximum gain for which the control system is stable. (b) The gain K must be negative for the system to be stable. (c) There exists a value of K for which the control system will oscillate at a frequency between 4 and 6 rad/sec. (d) The system is unstable if |K| > 10. (e) If K must be negative for stability, the control system cannot counteract a positive disturbance. (f) A positive constant disturbance will speed up the load, thereby making the final value of e negative. (g) With only a positive constant command input r, the error signal e must have a final value greater than zero. (h) For K = −1 the closed-loop system is stable, and the disturbance results in a speed error whose steady-state magnitude is less than 5 rad/sec.

Figure 10.90 Servomechanism for Problem 10.6

Disturbance (0.1 Nm) Amplifier R







e

K

Actuator 

1000 s  1000







Load 1000 s(s2  2s 25)

(rad/sec)

Tachometer 0.01

10.7

A stick balancer and its corresponding control block diagram are shown in Fig. 10.91. The control is a torque applied about the pivot. (a) Using root-locus techniques, design a compensator D(s) that will place the dominant roots at s = −5 ± 5j (corresponding to ωn = 7 rad/sec, ζ = 0.707). (b) Use Bode plotting techniques to design a compensator D(s) to meet the following specifications: • Steady-state θ displacement of less than 0.001 for a constant input torque Td = 1, • Phase margin ≥ 50◦ , • Closed-loop bandwidth ∼ = 7 rad/sec.

Problems Figure 10.91 Servomechanism for Problem 10.7

u

Td







Td

1 (s2  64)

745

u

D(s)

10.8

Consider the standard feedback system drawn in Fig. 10.92. (a) Suppose 2500 K . s(s + 25) Design a lead compensator so that the phase margin of the system is more than 45◦ ; the steady-state error due to a ramp should be less than or equal to 0.01. Using the plant transfer function from part (a), design a lead compensator so that the overshoot is less than 25% and the 1% settling time is less than 0.1 sec. Suppose K G(s) = s(1 + 0.1s)(1 + 0.2s) and let the performance specifications now be Kv = 100 and PM ≥ 40◦ . Is the lead compensation effective for this system? Find a lag compensator, and plot the root locus of the compensated system. Using G(s) from part (c), design a lag compensator such that the peak overshoot is less than 20% and Kv = 100. Repeat part (c) using a lead–lag compensator. Find the root locus of the compensated system in part (e), and compare your findings with those from part (c). G(s) =

(b) (c)

(d) (e) (f)

Figure 10.92 Block diagram of a standard feedback control system

R





e

Dc(s)

G(s)

Y



10.9

Consider the system in Fig. 10.92, where G(s) =

300 . s(s + 0.225)(s + 4)(s + 180)

The compensator Dc (s) is to be designed so that the closed-loop system satisfies the following specifications: • • •

Zero steady-state error for step inputs, PM = 55◦ , GM ≥ 6 db, Gain crossover frequency is not smaller than that of the uncompensated plant.

(a) What kind of compensation should be used and why? (b) Design a suitable compensator Dc (s) to meet the specifications. 10.10 We have discussed three design methods: the root-locus method of Evans, the frequency-response method of Bode, and the state-variable pole-assignment method. Explain which of these methods is best described by the following statements (if you feel more than one method fits a given statement equally well, say so and explain why):

746 Chapter 10 Control System Design: Principles and Case Studies (a) This method is the one most commonly used when the plant description must be obtained from experimental data. (b) This method provides the most direct control over dynamic response characteristics such as rise time, percent overshoot, and settling time. (c) This method lends itself most easily to an automated (computer) implementation. (d) This method provides the most direct control over the steady-state error constants Kp and Kv . (e) This method is most likely to lead to the least complex controller capable of meeting the dynamic and static accuracy specifications. (f) This method allows the designer to guarantee that the final design will be unconditionally stable. (g) This method can be used without modification for plants that include transportation lag terms—for example, e−2s G(s) = . (s + 3)2 10.11 Lead and lag networks are typically employed in designs based on frequency-response (Bode) methods. Assuming a Type 1 system, indicate the effect of these compensation networks on each of the listed performance specifications. In each case, indicate the effect as “an increase,” “substantially unchanged,” or “a decrease.” Use the secondorder plant G(s) = K/[s(s + 1)] to illustrate your conclusions. (a) (b) (c) (d) (e)

Kv Phase margin Closed-loop bandwidth Percent overshoot Settling time

10.12 Altitude Control of a Hot-Air Balloon: American solo balloonist Steve Fossett landed in the Australian outback aboard Spirit of Freedom on July 3rd, 2002, becoming the first solo balloonist to circumnavigate the globe (see Fig. 10.93). The equations of vertical motion for a hot-air balloon (Fig. 10.94), linearized about vertical equilibrium, are δ T˙ +

1 δT = δq, τ1

τ2 z¨

+ z˙ =

aδT + w,

where δT = deviation of the hot-air temperature from the equilibrium temperature where buoyant force equals weight, z = altitude of the balloon, δq = deviation in the burner heating rate from the equilibrium rate (normalized by the thermal capacity of the hot air), w = vertical component of wind velocity, τ1 , τ2 , a = parameters of the equations. An altitude-hold autopilot is to be designed for a balloon whose parameters are τ1 = 250 sec

τ2 = 25 sec a = 0.3 m/(sec·◦ C).

Problems Figure 10.93 Spirit of Freedom balloon Source: French Navy/Tahitipresse

Figure 10.94 Hot-air balloon

Hot air

Burner flame

w Wind z Ground

747

748 Chapter 10 Control System Design: Principles and Case Studies Only altitude is sensed, so a control law of the form δq(s) = D(s)[zd (s) − z(s)] will be used, where zd is the desired (commanded) altitude. (a) Sketch a root locus of the closed-loop eigenvalues with respect to the gain K for a proportional feedback controller, δq = −K(z − zd ). Use Routh’s criterion (or let s = jω and find the roots of the characteristic polynomial) to determine the value of the gain and the associated frequency at which the system is marginally stable. (b) Our intuition and the results of part (a) indicate that a relatively large amount of lead compensation is required to produce a satisfactory autopilot. Because Steve Fossett was a millionaire, he could afford a more complex controller implementation. Sketch a root locus of the closed-loop eigenvalues with respect to the gain K for a double-lead compensator, δq = D(s)(zd − z), where  D(s) = K

 s + 0.03 2 . s + 0.12

(c) Sketch the magnitude portions of the Bode plots (straight-line asymptotes only) for the open-loop transfer functions of the proportional feedback and lead-compensated systems. (d) Select a gain K for the lead-compensated system to give a crossover frequency of 0.06 rad/sec. (e) With the gain selected in part (d), what is the steady-state error in altitude for a steady vertical wind of 1 m/sec? (Be careful: First find the closed-loop transfer function from w to the error.) (f) If the error in part (e) is too large, how would you modify the compensation to give higher low-frequency gain? (Give a qualitative answer only.) 10.13 Satellite-attitude control systems often use a reaction wheel to provide angular motion. The equations of motion for such a system are Satellite : Wheel : Measurement : Control :

I φ¨ = Tc + Tex , J r˙ = −Tc , Z˙ = φ˙ − aZ, Tc = −D(s)(Z − Zd ),

where J = moment of inertia of the wheel, r = wheel speed, Tc = control torque, Tex = disturbance torque, φ = angle to be controlled, Z = measurement from the sensor, Zd = reference angle, I = satellite inertia (1000 kg/m2 ),

Problems

749

a = sensor constant (1 rad/sec), D(s) = compensation. (a) Suppose D(s) = K0 , a constant. Draw the root locus with respect to K0 for the resulting closed-loop system. (b) For what range of K0 is the closed-loop system stable? (c) Add a lead network with a pole at s = −1 so that the closed-loop system has a bandwidth ωBW = 0.04 rad/sec, a damping ratio ζ = 0.5, and compensation given by s+z . D(s) = K1 s+1 Where should the zero of the lead network be located? Draw the root locus of the compensated system, and give the value of K1 that allows the specifications to be met. (d) For what range of K1 is the system stable? (e) What is the steady-state error (the difference between Z and some reference input Zd ) to a constant disturbance torque Tex for the design of part (c)? (f) What is the type of this system with respect to rejection of Tex ? (g) Draw the Bode plot asymptotes of the open-loop system, with the gain adjusted for the value of K1 computed in part (c). Add the compensation of part (c), and compute the phase margin of the closed-loop system. ˙ and (h) Write state equations for the open-loop system, using the state variables φ, φ, Z. Select the gains of a state-feedback controller Tc = −Kφ φ − Kφ φ˙ to locate the √ closed-loop poles at s = −0.02 ± 0.02j 3. 10.14 Three alternative designs are sketched in Fig. 10.95 for the closed-loop control of a system with the plant transfer function G(s) = 1/s(s + 1). The signal w is the plant noise and may be analyzed as if it were a step; the signal v is the sensor noise and may be analyzed as if it contained power to very high frequencies. (a) Compute values for the parameters K1 , a, K2 , KT , K3 , d, and KD so that in each case (assuming w = 0 and v = 0), Y 16 . = 2 R s + 4s + 16 Note that in system III, a pole is to be placed at s = −4. (b) Complete the following table, expressing the last entries as A/sk to show how fast noise from v is attenuated at high frequencies:

System

Kv

y


w s=0

y


v s→∞

I II III

(c) Rank the three designs according to the following characteristics (the best as “1,” the poorest as “3”):

750 Chapter 10 Control System Design: Principles and Case Studies Figure 10.95 Alternative feedback structures for Problem 10.14

w I: r





K1



s1 sa



u 



1 s(s  1)

y 





@

w II: r





K2













u



1 s(s  1)

y 





@

KTs

w III: r







K3

s4 sd









u 



1 s(s  1)

y 

 KD

I

II



@

s sd

III

Tracking Plant-noise rejection Sensor-noise rejection 10.15 The equations of motion for a cart–stick balancer with state variables of stick angle, stick angular velocity, and cart velocity are ⎤ ⎤ ⎡ ⎡ 0 1 0 0 0 0.016 ⎦ x + ⎣ −0.649 ⎦ u, x˙ = ⎣ 31.33 −31.33 0 −0.216 8.649 y = [ 10

0

0 ]x,

where the output is stick angle, and the control input is voltage on the motor that drives the cart wheels. (a) Compute the transfer function from u to y, and determine the poles and zeros. (b) Determine the feedback gain K necessary to move the poles of the system to the locations −2.832 and −0.521 ± 1.068j, with ωn = 4 rad/sec. (c) Determine the estimator gain L needed to place the three estimator poles at −10.

Problems

751

(d) Determine the transfer function of the estimated-state-feedback compensator defined by the gains computed in parts (b) and (c). (e) Suppose we use a reduced-order estimator with poles at −10 and −10. What is the required estimator gain? (f) Repeat part (d) using the reduced-order estimator. (g) Compute the frequency response of the two compensators. 10.16 A 282-ton Boeing 747 is approaching land at sea level. If we use the state given in the case study (Section 10.3) and assume a velocity of 221 ft/sec (Mach 0.198), then the lateral-direction perturbation equations are ⎤ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎡ 0.0148 β −0.0890 −0.989 0.1478 0.1441 β˙ ⎥ ⎢ r ⎥ ⎢ −0.151 ⎥ ⎢ r˙ ⎥ ⎢ 0.168 −0.217 −0.166 0 ⎥ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎢ ⎦ ⎣ p ⎦ + ⎣ 0.0636 ⎦ δr, ⎣ p˙ ⎦ = ⎣ −1.33 0.327 −0.975 0 0 φ 0 0.149 1 0 φ˙ ⎤ ⎡ β ⎢ r ⎥ ⎥ y = [ 0 1 0 0 ]⎢ ⎣ p ⎦. φ The corresponding transfer function is G(s) =

−0.151(s + 1.05)(s + 0.0328 ± 0.414j) r(s) = . δr(s) (s + 1.109)(s + 0.0425)(s + 0.0646 ± 0.731j)

(a) Draw the uncompensated root locus [for 1 + KG(s)] and the frequency response of the system. What type of classical controller could be used for this system? (b) Try a state-variable design approach by drawing a SRL for the system. Choose the closed-loop poles of the system on the SRL to be αc (s) = (s + 1.12)(s + 0.165)(s + 0.162 ± 0.681j), and choose the estimator poles to be five times faster at αe (s) = (s + 5.58)(s + 0.825)(s + 0.812 ± 3.40j). (c) Compute the transfer function of the SRL compensator. (d) Discuss the robustness properties of the system with respect to parameter variations and unmodeled dynamics. (e) Note the similarity of this design to the one developed for different flight conditions earlier in the chapter. What does this suggest about providing a continuous (nonlinear) control throughout the operating envelope? 10.17 (Contributed by Prof. L. Swindlehurst) The feedback control system shown in Fig. 10.96 is proposed as a position control system. A key component of this system is an armature-controlled DC motor. The input potentiometer produces a voltage Ei that is proportional to the desired shaft position: Ei = Kp θi . Similarly, the output potentiometer produces a voltage E0 that is proportional to the actual shaft position: E0 = Kp θ0 . Note that we have assumed that both potentiometers have the same proportionality constant. The error signal Ei − E0 drives a compensator, which in turn produces an armature voltage that drives the motor. The motor has an armature resistance Ra , an armature inductance La , a torque constant Kt , and a back emf constant Ke . The moment of inertia of the motor shaft is Jm , and the rotational damping due to

752 Chapter 10 Control System Design: Principles and Case Studies Figure 10.96 A servomechanism with gears on the motor shaft and potentiometer sensors

Input potentiometer

JL

uo

ui

Eb

Output potentiometer

 Ei





E

Jm

ya

km

Dc

Compensator

Eb

Gears

Motor Eo

Ei
voltage from input potentiometer
Kpui Eo
voltage from output potentiometer
Kpuo E
Ei  Eo
error voltage
Kp(ui  uo) ya
armature voltage applied to motor Eb
battery voltage applied the potentiometers

bearing friction is Bm . Finally, the gear ratio is N : 1, the moment of inertia of the load is JL , and the load damping is BL . (a) Write the differential equations that describe the operation of this feedback system. (b) Find the transfer function relating θ0 (s) and θi (s) for a general compensator Dc (s). (c) The open-loop frequency-response data shown in Table 10.2 were taken using the armature voltage va of the motor as an input and the output potentiometer voltage E0 as the output. Assuming that the motor is linear and minimum-phase, make an estimate of the transfer function of the motor, G(s) =

θm (s) , Va (s)

where θm is the angular position of the motor shaft.

TABLE 10.2

Frequency-Response Data for Problem 10.8 Frequency (rad/sec) 0.1 0.2 0.3 0.5 0.8 1.0 2.0 3.0 4.0 5.0 7.0




E0 (s)



V (s)
(db) a

Frequency (rad/sec)

60.0 54.0 50.0 46.0 42.0 40.0 34.0 30.5 27.0 23.0 19.5

10.0 20.0 40.0 60.0 65.0 80.0 100.0 200.0 300.0 500.0




E0 (s)



V (s)
(db) a

14.0 2.0 −10.0 −20.0 −21.0 −24.0 −30.0 −48.0 −59.0 −72.0

Problems

753

(d) Determine a set of performance specifications that are appropriate for a position control system and will yield good performance. Design Dc (s) to meet these specifications. (e) Verify your design through analysis and simulation using MATLAB. 10.18 Design and construct a device to keep a ball centered on a freely swinging beam. An example of such a device is shown in Fig. 10.97. It uses coils surrounding permanent magnets as the actuator to move the beam, solar cells to sense the ball position, and a hall-effect device to sense the beam position. Research other possible actuators and sensors as part of your design effort. Compare the quality of the control achievable for ball-position feedback only with that of multiple-loop feedback of both ball and beam position.

Figure 10.97 Ball-balancer design example Source: Photo courtesy of David Powell

10.19 Design and construct the magnetic levitation device shown in Fig. 9.2. You may wish to use LEGO components in your design. 10.20 Run-to-Run Control: Consider the RTP system shown in Fig. 10.98. We wish to heat up a semiconductor wafer, and control the wafer surface temperature accurately using rings of tungsten halogen lamps. The output of the system is temperature T as a function of time: y = T (t). The system reference input R is a desired step in temperature (700◦ C), and the control input is lamp power. A pyrometer is used to measure the wafer center temperature. The model of the system is first order, and an integral controller is used as shown in Fig. 10.98. Normally, there is not a sensor bias (b = 0).

Figure 10.98 RTP system

 R 



K s

A sa

y
T 

  b

Ox Metrology

754 Chapter 10 Control System Design: Principles and Case Studies (a) Suppose the system suddenly develops a sensor bias b = 0, where b is known. What can be done to ensure zero steady-state tracking of temperature command R, despite the presence of the sensor bias? (b) Now assume b = 0. In reality, we are trying to control the thickness of the oxide film grown (Ox) on the wafer and not the temperature. At present, no sensor can measure Ox in real time. The semiconductor process engineer must use off-line equipment (called metrology) to measure the thickness of the oxide film grown on the wafer. The relationship between the system output temperature and Ox is nonlinear and given by

tf − c Oxide thickness = pe T (t) dt, 0

where tf is the process duration, and p and c are known constants. Suggest a scheme in which the center wafer oxide thickness Ox can be controlled to a desired value (say, Ox = 5000 Å) by employing the temperature controller and the output of the metrology. 10.21 Develop a nonlinear model for a tungsten halogen lamp and simulate it in SIMULINK. 10.22 Develop a nonlinear model for a pyrometer. Show how temperature can be deduced from the model. 10.23 Repeat the RTP case study design by summing the three sensors to form a single signal to control the average temperature. Demonstrate the performance of the linear design, and validate the performance on the nonlinear SIMULINK simulation. 10.24 One of the steps in semiconductor wafer manufacturing during photolithography is performed by placement of the wafer on a heated plate for a certain period of time. Laboratory experiments have shown that the transfer function from the heater power, u, to the wafer temperature, y, is given by y(s) 0.09 = G(s) = . u(s) (s + 0.19)(s + 0.78)(s + 0.00018) (a) Sketch the 180◦ root locus for the uncompensated system. (b) Using the root-locus design techniques, design a dynamic compensator, D(s), such that the system meets the following time-domain specifications i. ii. iii. iv

Mp ≤ 5% tr ≤ 20 sec ts ≤ 60 sec Steady-state error to a 1◦ C step input command < 0.1◦ C. Draw the 180◦ root locus for the compensated system.

10.25 Excitation-Inhibition Model from Systems Biology (Yang and Iglesias, 2005): In Dictyostelium cells, the activation of key signaling molecules involved in chemoattractant sensing can be modeled by the following third order linearized model. The external disturbance to the output transfer function is: (1 − α)s y(s) = S(s) = w(s) (s + α)(s + 1)(s + γ ) where, w is the external disturbance signal proportional to chemoattractant concentration, and y is the output which is the fraction of active response regulators. Show that there is an alternate representation of the system with the “plant” transfer function (1 − α) G(s) = 2 s + (1 + α + γ )s + (α + γ + αγ )

Problems

755

and the “feedback regulator” D(s) =

αγ . (1 − α)s

It is known that α = 1 for this version of the model. Draw the feedback block diagram of the system showing the locations of the disturbance input and the output. What is the significance of this particular representation of the system? What hidden system property does it reveal? Is the disturbance rejection a robust property for this system? Plot the disturbance rejection response of the system for a unit step disturbance input. Assume the system parameter values are α = 0.5 and γ = 0.2.

This page intentionally left blank

Appendix A Laplace Transforms A.1 The L− Laplace Transform Laplace transforms can be used to study the complete response characteristics of feedback systems, including the transient response. This is in contrast to Fourier transforms, in which the steady-state response is the main concern. In many applications it is useful to define the Laplace transform of f (t), denoted by L− {f (t)} = F(s), as a function of the complex variable s = σ + jω, where

∞ F(s)  f (t)e−st dt, (A.1) 0−

which uses 0− (that is, a value just before t = 0) as the lower limit of integration and is referred to as the unilateral (or one-sided) Laplace transform.1 A function f (t) will have a Laplace transform if it is of exponential order, which means that there exists a real number σ such that (A.2) lim |f (t)e−σ t | = 0. t→∞

The decaying exponential term in the integrand in effect provides a built-in convergence factor. This means that even if f (t) does not vanish as t → ∞, the integrand will vanish for sufficiently large values of σ if f does not grow at a faster-than-exponential 2 rate. For example, aebt is of exponential order, whereas et is not. If F(s) exists for some s0 = σ0 + jω0 , then it exists for all values of s such that (A.3) Re(s) ≥ σ0 . The smallest value of σ0 for which F(s) exists is called the abscissa of convergence, and the region to the right of Re(s) ≥ σ0 is called the region of convergence. Typically, two-sided Laplace transforms exist for a specified range α < Re(s) < β, (A.4) which defines the strip of convergence. Table A.2 gives some Laplace transform pairs. Each entry in the table follows from direct application of the transform definition.2

A.1.1 Properties of Laplace Transforms In this section we will address and prove each of the significant properties of the Laplace transform as discussed in Chapter 3 and listed in Table A.1. In addition we show how these properties can be used through examples. 1 Bilateral (or two-sided) Laplace transforms and the so-called L transforms, in which the lower value + of integral is 0+ , also arise elsewhere. 2As for the one-sided Laplace transform, an astute reader would wonder what happens to the validity of the Laplace transform for the rest of the s-plane, namely, the region where Re(s) < σ0 . Indeed it would be disappointing if F(s) was only valid for Re(s) ≥ σ0 and not elsewhere in the s-plane. Fortunately, except for some pathological cases (which do not arise in practice), one can invoke an important result from the theory of complex variables known as the Analytic Continuation Theorem to extend the region of the validity of F(s) to the whole s-plane excluding the locations of the poles.

757

758 Appendix A Laplace Transforms

TABLE A.1

Properties of Laplace Transforms Number

Laplace Transform

—

Time Function

Comment

F (s)

f (t)

Transform pair

1

αF1 (s) + βF2 (s)

αf1 (t) + βf2 (t)

Superposition

2

f (t − λ)

Time delay (λ ≥ 0)

f (at)

Time scaling

4

F (s)e−sλ 1
s F |a| a F (s + a)

e−at f (t)

Shift in frequency

5

sm F (s) − sm−1 f (0)

f (m) (t)

t f (ζ ) dζ

Differentiation Integration

f1 (t) ∗ f2 (t)

Convolution

f (0+ )

Initial Value Theorem

3

−sm−2 f˙ (0) − · · · − f (m−1) (0)

7

1 F (s) s F1 (s)F2 (s)

8

lim sF (s)

6

9 10 11 12

0

s→∞

lim sF (s)

lim f (t)

s→0

σc +j∞ 1 F1 (ζ )F2 (s − ζ )dζ 2π j σc −j∞

+j∞ 1 Y (−jω)U(jω) dω 2π −j∞ d − F (s) ds

Final Value Theorem

t→∞

f1 (t)f2 (t)

Time product

∞

Parseval’s Theorem

0

y(t)u(t) dt

tf (t)

Multiplication by time

1. Superposition One of the more important properties of the Laplace transform is that it is linear. We can prove this as follows:

∞ L{αf1 (t) + βf2 (t)} = [αf1 (t) + βf2 (t)]e−st dt (A.5) 0



=α

∞

f1 (t)e−st dt + β



0

∞

f2 (t)e−st dt

0

= αF1 (s) + βF2 (s). The scaling property is a special case of linearity; that is, L{αf (t)} = αF(s).

EXAMPLE A.1

Sinusoidal Signal Find the Laplace transform of f (t) = 1 + 2 sin(ωt). Solution. The Laplace transform of sin(ωt) is ω L{sin(ωt)} = 2 . s + ω2

(A.6)

A.1 The L− Laplace Transform

TABLE A.2

Table of Laplace Transforms F (s)

f (t), t ≥ 0

1

1

δ(t)

2

1/s

1(t)

3

1/s2

t

4

2!/s3

t2

5

3!/s4

t3

6

m!/sm+1

tm

1 s+a 1 (s + a)2 1 (s + a)3 1 (s + a)m a s(s + a) a s2 (s + a) b−a (s + a)(s + b) s (s + a)2 a2 s(s + a)2 (b − a)s (s + a)(s + b) a s 2 + a2 s s 2 + a2 s+a (s + a)2 + b2 b (s + a)2 + b2 a 2 + b2 s[(s + a)2 + b2 ]

e−at

Number

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

te−at 1 2 −at t e 2! 1 t m−1 e−at (m − 1)! 1 − e−at 1 (at − 1 + e−at ) a e−at − e−bt (1 − at)e−at 1 − e−at (1 + at) be−bt − ae−at sin at cos at e−at cos bt e−at sin bt 
a 1 − e−at cos bt + sin bt b

Therefore, using Eq. (A.5) we obtain 2ω 1 s2 + 2ωs + ω2 + 2 . = s s + ω2 s3 + ω 2 s The following commands in MATLAB
yields the same result, syms s t w laplace(1+2*sin(w*t)). F(s) =

759

760 Appendix A Laplace Transforms 2. Time Delay Suppose a function f (t) is delayed by λ > 0 units of time. Its Laplace transform is

∞ F1 (s) = f (t − λ)e−st dt. 0

Let us define t
= t − λ. Then dt
= dt, because λ is a constant and f (t) = 0 for t < 0. Thus

∞

∞

f (t
)e−s(t +λ) dt
= f (t
)e−s(t +λ) dt
. F1 (s) = −λ

Because

e−sλ

0

is independent of time, it can be taken out of the integrand, so

∞
f (t
)e−st dt
= e−sλ F(s). F1 (s) = e−sλ

(A.7)

0

From this result we see that a time delay of λ corresponds to multiplication of the transform by e−sλ .

EXAMPLE A.2

Delayed Sinusoidal Signal Find the Laplace transform of f (t) = A sin(t − td ). Solution. The Laplace transform of sin(t) is L{sin(t)} =

s2

1 . +1

Therefore, using Eq. (A.7) we obtain F(s) =

A e−std . s2 + 1

3. Time Scaling If the time t is scaled by a factor a, then the Laplace transform of the time-scaled signal is

∞ F1 (s) = f (at)e−st dt. 0

Again, we define t
= at. As before, dt
= a dt and

∞
e−st /a
1
s dt = F . f (t
) F1 (s) = |a| |a| a 0

EXAMPLE A.3

Sinusoid with Frequency ω Find the Laplace transform of f (t) = A sin(ωt). Solution. The Laplace transform of sin(t) is L{sin(t)} =

1 . s2 + 1

(A.8)

A.1 The L− Laplace Transform

761

Therefore, using Eq. (A.8) we obtain 1 1 * s +2 |ω| ω +1 Aω , = 2 s + ω2

F(s) =

as expected. The following commands in MATLAB yields the same result, syms s t w A laplace(A*sin(w*t)).

4. Shift in Frequency Multiplication (modulation) of f (t) by an exponential expression in the time domain corresponds to a shift in frequency:

∞

∞ −at −st e f (t)e dt = f (t)e−(s+a)t dt = F(s + a). (A.9) F1 (s) = 0

EXAMPLE A.4

0

Exponentially Decaying Sinusoid Find the Laplace transform of f (t) = A sin(ωt)e−at . Solution. The Laplace transform of sin(ωt) is L{sin(ωt)} =

s2

ω . + ω2

Therefore, using Eq. (A.9) we obtain F(s) =

Aω . (s + a)2 + ω2

5. Differentiation The transform of the derivative of a signal is related to its Laplace transform and its initial condition as follows:   ∞ 

∞ df df + s f (t)e−st dt. (A.10) = e−st dt = e−st f (t)|∞ L − 0 dt dt 0− 0− Because f (t) is assumed to have a Laplace transform, e−st f (t) → 0 as t → ∞. Thus L[f˙ ] = −f (0− ) + sF(s).

(A.11)

Another application of Eq. (A.11) leads to L{f¨ } = s2 F(s) − sf (0− ) − f˙ (0− ).

(A.12)

762 Appendix A Laplace Transforms Repeated application of Eq. (A.11) leads to L{f m (t)} = sm F(s) − sm−1 f (0− ) − sm−2 f˙ (0− ) − · · · − f (m−1) (0− ),

(A.13)

where f m (t) denotes the mth derivative of f (t) with respect to time.

EXAMPLE A.5

Derivative of Cosine Signal Find the Laplace transform of g(t) = d f (t), where f (t) = cos(ωt). dt Solution. The Laplace transform of cos(ωt) is s . F(s) = L{cos(ωt)} = 2 s + ω2 Using Eq. (A.11) with f (0− ) = 1, we have G(s) = L{g(t)} = s ·

s2

ω2 s −1=− 2 . 2 +ω s + ω2

6. Integration Let us assume that we wish to determine the Laplace transform of the integral of a time function—that is, to find  t  ∞  t  f (ξ ) dξ = f (ξ ) dξ e−st dt. F1 (s) = L 0

0

Employing integration by parts, where

t u= f (ξ ) dξ

and

0

dν = e−st dt,

0

we get 

1 F1 (s) = − e−st s

EXAMPLE A.6



t

f (ξ ) dξ

0

∞



∞

− 0

0

1 1 − e−st f (t) dt = F(s). s s

Time Integral of Sinusoidal Signal t Find the Laplace transform of f (t) = 0 sin ωτ dτ . Solution. The Laplace transform of sin(ωt) is ω . L{sin(ωt)} = 2 s + ω2 Therefore, using Eq. (A.14), then F(s) =

s3

ω . + ω2 s

(A.14)

A.1 The L− Laplace Transform

763

7. Convolution Convolution in the time domain corresponds to multiplication in the frequency domain. Assume that L{f1 (t)} = F1 (s) and L{f2 (t)} = F2 (s). Then 

∞

∞  t −st L{f1 (t) ∗ f2 (t)} = f1 (t) ∗ f2 (t)e dt = f1 (τ )f2 (t − τ ) dτ e−st dt. 0

0

0

We see that t varies from zero to infinity and τ varies from zero to t. With the aid of Fig. A.1, we reverse the order of integration and change the limits of integration accordingly so that τ varies from zero to infinity and ∞  t  τ , to yield

∞ ∞ f1 (τ )f2 (t − τ )e−st dt dτ . L{f1 (t) ∗ f2 (t)} = 0

τ

Multiplying by e−sτ esτ results in 

∞ −sτ L{f1 (t) ∗ f2 (t)} = f1 (τ )e 0

∞ τ

f2 (t − τ )e

If we change variables t
 t − τ , then

∞

f1 (τ )e−st dτ L{f1 (t) ∗ f2 (t)} = 0

∞

−s(t−τ )

 dt dτ .




f2 (t
)e−st dt
,

0

L{f1 (t) ∗ f2 (t)} = F1 (s)F2 (s). This implies that L−1 {F1 (s)F2 (s)} = f1 (t) ∗ f2 (t).

EXAMPLE A.7

(A.15)

Ramp Response of a First-Order System Find the ramp response of a first-order system with a pole at +a. Solution. Let f1 (t) = t be the ramp input and f2 (t) = eat be the impulse response of the first-order system. Then, using Eq. (A.15) we find that   1 −1 1 L = f1 (t) ∗ f2 (t) s2 s − a

t f1 (τ )f2 (t − τ ) dτ = 0

=

t

τ ea(t−τ ) dτ

0

=

1 at (e − at − 1). a2

The following commands in MATLAB yields the same result, syms s t a ilaplace(1/(sˆ3−a*sˆ2)).

764 Appendix A Laplace Transforms Figure A.1 Diagram illustrating reversal of the order of integration

t

t=τ

0

τ

τ

8. Time Product Multiplication in the time domain corresponds to convolution in the frequency domain: 1 2πj

L{f1 (t)f2 (t)} =



σc +j∞ σc −j∞

F1 (ξ )F2 (s − ξ ) dξ .

To see this, consider the relation

∞

L{f1 (t)f2 (t)} =

f1 (t)f2 (t)e−st dt.

0

Substituting the expression for f1 (t) given by Eq. (3.25) yields ∞

L{f1 (t)f2 (t)} = 0

1 2πj



σc +j∞

σc −j∞

ξt



F1 (ξ )e dξ f2 (t)e−st dt.

Changing the order of integration results in

L{f1 (t)f2 (t)} =

1 2πj



σc +j∞ σc −j∞



∞

F1 (ξ ) 0

f2 (t)e−(s−ξ )t dt dξ .

A.1 The L− Laplace Transform

765

Using Eq. (A.9), we get L{f1 (t)f2 (t)} =

1 2π j



σc +j∞ σc −j∞

F1 (ξ )F2 (s − ξ ) dξ =

1 F1 (s) ∗ F2 (s). 2πj

(A.16)

9. Parseval’s Theorem Parseval’s famous theorem is used to compute the “energy” in a signal or “correlation” between two signals. It tells us that mentioned quantities can be computed either in the time domain or in the frequency domain. If

∞

∞ |y(t)|2 dt < 1 and |u(t)|2 dt < 1 (A.17) 0

0

(i.e., y(t) and u(t) are square integrable), then

∞

∞ 1 y(t)u(t)dt = Y (−jω)U(jω) dω. 2π −∞ 0

(A.18)

Parseval’s result involves only a substitution of the transform for the time functions and an exchange of integration:  

∞

∞

∞ 1 jωt y(t)u(t) dt = y(t) U(jω)e dω dt (A.19) 2π −∞ 0 0  ∞ 

∞ 1 U(jω) y(t)e jωt dt dω (A.20) = 2π −∞ 0

∞ 1 = U(jω)Y (−jω) dω. (A.21) 2π −∞

10. Multiplication by Time Multiplication by time corresponds to differentiation in the frequency domain. Let us consider

d d ∞ −st F(s) = e f (t) dt ds ds 0

∞ = −te−st f (t) dt 0



∞

=−

e−st [tf (t)] dt

0

= −L{tf (t)}. Then L{tf (t)} = −

d F(s), ds

which is the desired result.

EXAMPLE A.8

Time Product of Sinusoidal Signal Find the Laplace transform of f (t) = t sin ωt.

(A.22)

766 Appendix A Laplace Transforms Solution. The Laplace transform of sin ωt is L{sin(ωt)} =

s2

ω . + ω2

Hence, using Eq. (A.22), we obtain   2ωs ω d = 2 F(s) = − . 2 2 ds s + ω (s + ω2 )2 The following commands in MATLAB yields the same result, syms s t w laplace(t*sin(w*t)).

A.1.2 Inverse Laplace Transform by Partial-Fraction Expansion As we saw in Chapter 3, the easiest way to find f (t) from its Laplace transform F(s), if F(s) is rational, is to expand F(s) as a sum of simpler terms that can be found in the tables via partial-fraction expansion. We have already discussed this method in connection with simple roots in Section 3.1.5. In this section, we discuss partial-fraction expansion for cases of complex and repeated roots. Complex Poles In the case of quadratic factors in the denominator, the numerator of the quadratic factor is chosen to be first order as shown in Example A.9. Whenever there exists a complex conjugate pair of poles such as C2 C1 + , F(s) = s − p1 s − p∗1 we can show that

C2 = C1∗

(see Problem 3.1) and that ∗t

f (t) = C1 ep1 t + C1∗ ep1 = 2Re(C1 ep1 t ). Assuming that p1 = α + jβ, we may rewrite f (t) in a more compact form as f (t) = 2Re{C1 ep1 t } = 2Re{|C1 |ej arg(C1 ) e(α+jβ)t } αt

= 2|C1 |e cos[βt + arg(C1 )].

EXAMPLE A.9

Partial-Fraction Expansion: Distinct Complex Roots Find the function f (t) for which the Laplace transform is 1 . F(s) = s(s2 + s + 1) Solution. We rewrite F(s) as C2 s + C 3 C1 + 2 . s s +s+1 Using the cover-up method, we find that F(s) =

C1 = sF(s)|s=0 = 1.

(A.23)

A.1 The L− Laplace Transform

767

Setting C1 = 1 and then equating the numerators in the partial-fraction expansion relation, we obtain (s2 + s + 1) + (C2 s + C3 )s = 1. After solving for C2 and C3 , we find that C2 = −1 and C3 = −1. To make it more suitable for using the Laplace transform tables, we rewrite the partial fraction as F(s) = From the tables we have, 

s + 21 + 21 1 . −
2 s s + 21 + 43

  1 3 3 f (t) = 1 − e−t/2 cos t − √ e−t/2 sin t 1(t) 4 4 3 √   3 π 2 −t/2 t− cos = 1− √ e 1(t). 2 6 3 

Alternatively, we may write F(s) as F(s) =

C2∗ C2 C1 + + , s s − p1 s − p∗1

(A.24)

√ where p1 = − 21 + j 23 . C1 = 1, as before, and now 1 1 C2 = (s − p1 )F(s)|s=p1 = − + j √ , 2 2 3 1 1 C2∗ = − − j √ , 2 2 3 and f (t) = (1 + 2|C2 |eαt cos[βt + arg(C2 )])1(t)  √  5π 3 2 −t/2 t+ = 1+ √ e cos 1(t) 2 6 3 √   5π 3 2 −t/2 t− cos 1(t). = 1− √ e 2 6 3 The latter partial-fraction expansion can be readily computed using the MATLAB statements num = 1;

% form numerator

den = conv([1 0],[1 1 1]);

% form denominator

[r,p,k] = residue(num,den)

% compute residues

which yield the results r = [−0.5000 + 0.2887i − 0.5000 − 0.2887i 1.0000]
; p = [−0.5000 + 0.8660i − 0.5000 − 0.8660i 0]
; k = []

768 Appendix A Laplace Transforms and agrees with the previous hand calculations. Note that if we are using the tables, the first method is preferable, while the second method is preferable for checking MATLAB results. The following commands in MATLAB yield the same result for the inverse Laplace transform, syms s t ilaplace(1/(s*(sˆ2+s+1))). Repeated poles For the case in which F(s) has repeated roots, the procedure to compute the partial-fraction expansion must be modified. If p1 is repeated three times, we write the partial fraction as C1 C2 C3 C4 Cn F(s) = + + + + ··· + . 2 3 s − p1 (s − p1 ) (s − p1 ) s − p4 s − pn We determine the constants C4 through Cn as discussed previously. If we multiply both sides of the preceding equation by (s − p1 )3 , we obtain Cn (s − p1 )3 . (A.25) s − pn If we then set s = p1 , all the factors on the right side of Eq. (A.25) will go to zero except C3 , which is C3 = (s − p1 )3 F(s)|s=p1 , as before. To determine the other factors, we differentiate Eq. (A.25) with respect to the Laplace variable s:   d d Cn (s − p1 )3 [(s − p1 )3 F(s)] = 2C1 (s − p1 ) + C2 + · · · + . (A.26) ds ds s − pn Again, if we set s = p1 , we have d C2 = [(s − p1 )3 F(s)]s=p1 . ds Similarly, if we differentiate Eq. (A.26) again and set s = p1 a second time, we get (s − p1 )3 F(s) = C1 (s − p1 )2 + C2 (s − p1 ) + C3 + · · · +

1 d2 [(s − p1 )3 F(s)]s=p1 . 2 ds2 In general, we may compute Ci for a factor with multiplicity k as   1 di k [(s − p ) F(s)] , i = 0, . . . , k − 1. Ck−i = 1 i! dsi s=p1 C1 =

EXAMPLE A.10

Partial-Fraction Expansion: Repeated Real Roots Find the function f (t) that has the Laplace transform s+3 . F(s) = (s + 1)(s + 2)2 Solution. We write the partial fraction as C2 C3 C1 + + . F(s) = s + 1 s + 2 (s + 2)2

A.1 The L− Laplace Transform

769

Then C1 = (s + 1)F(s)|s=−1 =

s+3 |s=−1 = 2, (s + 2)2

d  (s + 2)2 F(s) |s=−2 = −2, ds s+3 |s=−2 = −1. C3 = (s + 2)2 F(s)|s=−2 = s+1

C2 =

The function f (t) is

f (t) = (2e−t − 2e−2t − te−2t )1(t).

The partial fraction computation can also be carried out using MATLAB’s residue function, num = [1 3]; den = conv([1 1],[1 4 4]); [r,p,k] = residue(num,den)

% form numerator % form denominator % compute residues

which yields the result r = [−2 −1 2]’, p = [−2 −2 −1]’, and k = [ ]; and agrees with the hand calculations. The following commands in MATLAB yields the same result for the inverse Laplace transform, syms s t ilaplace((s+3)/((s+1)*(s+2)ˆ2)).

A.1.3 The Initial Value Theorem We discussed the Final Value Theorem in Chapter 3. A second valuable Laplace transform theorem is the Initial Value Theorem, which states that it is always possible to determine the initial value of the time function f (t) from its Laplace transform. We may also state the theorem in this way: The Initial Value Theorem

For any Laplace transform pair, lim sF(s) = f (0+ ).

s→∞

We may show this as follows. Using Eq. (A.11), we get  

∞ df df − = sF(s) − f (0 ) = L e−st dt. − dt dt 0 Let us consider the case in which s → ∞ and rewrite the integral as

∞

0+

∞ df (t) df (t) df (t) dt = dt + dt. e−st e−st e−st − + − dt dt dt 0 0 0

(A.27)

(A.28)

770 Appendix A Laplace Transforms Taking the limit of Eq. (A.28) as s → ∞, we get   +

∞ 0 − 0 df (t) −st df (t) dt + dt . lim [sF(s) − f (0 )] = lim e e s→∞ s→∞ dt dt 0+ 0− The second term on the right side of the preceding equation approaches zero as s → ∞, because e−st → 0. Hence lim [sF(s) − f (0− )] = lim [f (0+ ) − f (0− )] = f (0+ ) − f (0− )

s→∞

s→∞

or

lim sF(s) = f (0+ ).

s→∞

In contrast with the Final Value Theorem, the Initial Value Theorem can be applied to any function F(s).

EXAMPLE A.11

Initial Value Theorem Find the initial value of the signal in Example 3.11. Solution. From the Initial Value Theorem, we get 3 = 0, s(s − 2) which checks with the expression for y(t) computed in Example 3.11. y(0+ ) = lim sY (s) = lim s s→∞

s→∞

A.1.4 Final Value Theorem The Final Value Theorem

If all poles of sY (s) are in the left half of the s-plane, then lim y(t) = lim sY (s).

t→∞

s→0

(3.46)

Proof of the Final Value Theorem We may prove this result as follows. The derivative relationship developed in Eq. (3.33) is  

∞ dy dy e−st dt. = sY (s) − y(0− ) = L dt dt 0− We assume we are interested in the case where s → 0. Then  ∞  dy lim [sY (s) − y(0)] = lim e−st dt = lim [y(t) − y(0)], t→∞ s→0 s→0 dt 0 and we have lim y(t) = lim sY (s). t→∞

s→0

Another way to see this same result is to note that the partial-fraction expansion of Y (s) [Eq. (3.43)] is C2 Cn C1 + + ··· + . Y (s) = s − p1 s − p2 s − pn

A.1 The L− Laplace Transform

771

Let us say that p1 = 0 and all other pi are in the LHP so that C1 is the steady-state value of y(t). Using Eq. (3.45), we see that C1 = lim y(t) = sY (s)|s=0 , t→∞

which is the same as the previous result. For a thorough study of Laplace transforms and extensive tables, see Churchill (1972) and Campbell and Foster (1948); for the two-sided transform, see Van der Pol and Bremmer (1955).

Appendix B Solutions to the Review Questions Chapter 1 1. What are the main components of a feedback control system? The process, the actuator, the sensor, and the controller. 2. What is the purpose of the sensor? To measure the output variable and, usually, to convert it to an electrical voltage. 3. Give three important properties of a good sensor. A good sensor is linear (the output is proportional to the input signal) over a large range of amplitudes and a large range of frequencies at its input, has low noise, is unbiased, is easy to calibrate, and has low cost. The relative values of these properties varies with the particular application. 4. What is the purpose of the actuator? The actuator takes an input, usually electrical, and converts it to a signal such as a force or torque that causes the process output to move or change over the required range. 5. Give three important properties of a good actuator. A good actuator has fast response, and adequate power, energy, speed, torque, and so on, to be able to cause the process output to meet the design specifications and is efficient, lightweight, small, cheap, and so on. As with sensors, the relative value of these properties varies with the application. 6. What is the purpose of the controller? Give the input(s) and output(s) of the controller. The controller is to take the sensor output (the input to the controller) and compute the control signal (the output of the controller) to be sent to the actuator. 7. What physical variable(s) of a process can be directly measured by a Hall effect sensor? A Hall-effect device measures the strength of a magnetic field and can be most easily configured to measure relative positions of two bodies or relative angles. 8. What physical variable is measured by a tachometer? A tachometer measures speed of rotation or angular velocity. 9. Describe three different techniques for measuring temperature. In each of the following cases, it is important to realize that the devices mentioned need to be calibrated and often corrected for nonlinearity in order to give a reliable, accurate measure of temperature.

772

Appendix B Solutions to the Review Questions

773

(a) An early technique still used in many home thermostats is based on the bimetallic strip composed of two strips of different metals that expand with different coefficients with temperature. As a result, the strip bends with temperature and the resulting motion can be used as a measure of temperature. This principle was introduced in the 18th century to maintain a constant length to a clock pendulum for precision time keeping. (b) A technique related to the bimetallic strip is based on the fact that metals with different work functions placed in contact will produce a voltage that is proportional to temperature. Such a device is called a thermocouple and is the basis of a standard laboratory technique for measuring temperature. (c) A number of materials have electrical resistance that is dependent in a monotonic way on temperature, and a resistance bridge can be used with one of these to indicate temperature. Such devices are called thermistors. (d) For high temperatures, it is well known that the color of the radiation due to heat depends on temperature. A piece of iron placed in a fire will glow orange, then red, and finally become white hot at high temperatures. An instrument for measuring the frequency of the radiation, and thus the temperature, is a pyrometer. (e) In ceramic kilns, cones of different materials that melt at different and known temperatures are placed near the products in the kiln to indicate when the design temperature has been reached. The potter watches until the cone of importance begins to sag and then knows that the products should be removed. These give a quantized measure of temperature. 10. Why do most sensors have an electrical output, regardless of the physical nature of the variable being measured? Electrical signals are the most easily manipulated; therefore, most controllers are electrical devices, either analog or digital. To provide the signal input to such a device, the sensor needs to produce an electrical output.

Chapter 2 1. What is a “free-body” diagram? To write the equations of motion of a system of connected bodies, it is useful to draw each body in turn with the influence of all other bodies represented by forces and torques on the body in question. A drawing of the collection of such isolated bodies is called a “free-body diagram.” 2. What are the two forms of Newton’s law? Translational motion is described by F = ma. Rotational motion is described by M = Iα. 3. For a structural process to be controlled, such as a robot arm, what is the meaning of “collocated control”? “Noncollocated control”? When the actuator and the sensor are located on the same rigid body, the control is said to be “collocated.” When they are on different bodies that are connected by springs, the control is “noncollocated.”

774 Appendix B Solutions to the Review Questions 4. State Kirchhoff’s current law. The algebraic sum of all currents entering a junction or circuit is zero. 5. State Kirchhoff’s voltage law. The algebraic sum of voltages around a closed path in an electric circuit is zero. 6. When, why, and by whom was the device named an “operational amplifier”? In a paper in 1947, Ragazzini, Randall, and Russell named the high-gain, wide-bandwidth amplifier used in feedback to realize operational calculus “operations” the operational amplifier. 7. What is the major benefit of having zero input current to an operational amplifier? With zero input current the amplifier does not load the input circuit; thus, the transfer function of the device is not dependent of the amplifier characteristics. Also, the analysis of the circuit is simplified in this case. 8. Why is it important to have a small value for the armature resistance Ra of an electric motor? The armature resistance causes power loss when the armature current flows and thus reduces the efficiency of the motor. 9. What are the definition and units of the electric constant of a motor? A rotating motor produces a voltage (called the back emf) in its armature proportional to the rotational speed. The electric constant Ke is the ratio of this ˙ The units are volt-sec/radians. voltage to the speed, so that e = Ke θ. 10. What are the definition and units of the torque constant of an electric motor? When current ia flows in the armature of an electrical motor, a torque τ is produced that is proportional to the current. The torque constant Kt is the constant of proportionality, so that τ = Kt ia . The units are Newton-meters/amp. 11. Why do we approximate a physical model of the plant (which is always nonlinear) with a linear model? Analysis and design of linear models is vastly simpler than with nonlinear models. Furthermore, it has been shown (by Lyapunov) that, if the linear approximation is stable, then there is at least some region of stability for the nonlinear model. 

12. Give the relationships for (a) heat flow across a substance, and (b) heat storage in a substance. (a) Heat flow is proportional to the temperature difference divided by the thermal resistance; that is, 1 (T1 − T2 ). R (b) The differential equation describing the heat storage is q=

1 q, C where C is the thermal capacity of the material. T˙ =

Appendix B Solutions to the Review Questions 

775

13. Name and give the equations for the three relationships governing fluid flow. Continuity:

m ˙ = win − wout .

f = pA. 1 Resistance: w = (p1 − p2 )1/α . R

Force equilibrium:

Chapter 3 1. What is the definition of “transfer function”? The Laplace transform of the output of a linear, time-invariant system, Y (s), is proportional to the transform of its input, U(s). The function of proportionality is the transfer function F(s), so that Y (s) = F(s)U(s). It is assumed that all initial conditions are zero. 2. What are the properties of systems whose responses can be described by transfer functions? The system must be both linear (superposition applies) and time invariant (the parameters do not vary with time). 3. What is the Laplace transform of f (t − λ)1(t − λ) if the transform of f (t) is F(s)? L{f (t − λ)1(t − λ)} = e−sλ F(s). 4. State the Final Value Theorem. If all the poles of sF(s) are in the LHP, then the final value of f (t) is given by lim f (t) = lim sF(s). t→∞

s→0

5. What is the most common use of the Final Value Theorem in control? A standard test of a control system is the step response, and the FVT is used to determine the steady-state error to such an input. 6. Given a second-order transfer function with damping ratio ζ and natural frequency ωn , what is the estimate of the step response rise time? What is the estimate of the percent overshoot in the step response? What is the estimate of the settling time? These are given by tr ∼ = 1.8/ωn , Mp is set by the damping ratio (see the curve in Fig. 3.23) and ts ∼ = 4.6/σ . 7. What is the major effect of an extra zero in the LHP on the second-order step response? Such a zero causes additional overshoot, and the closer the zero is to the imaginary axis, the higher the overshoot. If the zero is more than six times the real part of the complex poles, the effect is negligible. 8. What is the most noticeable effect of a zero in the RHP on the step response of the second-order system? Such a zero often causes an initial undershoot of the response.

776 Appendix B Solutions to the Review Questions 9. What is the main effect of an extra real pole on the second-order step response? A pole slows down the response and makes the rise time longer. The closer the pole is to the imaginary axis, the more pronounced is the effect. If the pole is more than six times the real part of the complex poles, the effect is negligible. 10. Why is stability an important consideration in control system design? Almost any useful dynamic system must be stable to perform its function. Feedback around a system that is normally stable can actually introduce instability, so control designers must be able to assure the stability of their designs. 11. What is the main use of Routh’s criterion? With this method, we can find (symbolically) the range of a parameter such as the loop gain for which the system will be stable. 12. Under what conditions might it be important to know how to estimate a transfer function from experimental data? In many cases, the equations of motion are either extremely complex or not known at all. Chemical processes such as a paper-making machine are often of this kind. In these cases, if one wishes to build a good control, it is very useful to be able to take transient data or steady-state frequency-response data and to estimate a transfer function from these.

Chapter 4 1. Give three advantages of feedback in control. (a) Feedback can reduce the steady-state error in response to disturbances. (b) Feedback can reduce the steady-state error in tracking a reference. (c) Feedback can reduce the sensitivity of a transfer function to parameter changes. (d) Feedback can stabilize an unstable process. 2. Give two disadvantages of feedback in control. (a) Feedback requires a sensor, which can be very expensive and may introduce additional noise. (b) Feedback systems are often more difficult to design and operate than openloop systems. 3. What is the main objective of introducing integral control? Integral control will make the error to a constant input go to zero. It removes the effects of process noise bias. It cannot remove the effects of sensor bias. 4. What is the major objective of adding derivative control? Derivative control typically makes the system better damped and more stable. 5. Why might a designer wish to put the derivative term in the feedback rather than in the error path?

Appendix B Solutions to the Review Questions

777

When a reference input might include sudden changes, including it in the derivative action might cause unnecessary large controls. 6. A temperature control system is found to have zero error to a constant tracking input and an error of 0.5◦ C to a tracking input that is linear in time, rising at the rate of 40◦ C/sec. What is the system type of this control system and what is the relevant error constant (Kp or Kv or Ka )? The system is Type 1 and the Kv is the ratio of input rate to error or Kv = 40/0.5 = 80/ sec. 7. What are the units of Kp , Kv , and Ka ? Kp is dimensionless, Kv is sec−1 , and Ka is sec−2 . 8. What is the definition of system type with respect to reference inputs? With only a polynomial of degree k reference input (no disturbances), the type is the largest value of k for which the steady-state error is a constant. 9. What is the definition of system type with respect to disturbance inputs? With only a polynomial of degree k disturbance input (no reference), the type is the largest value of k for which the steady-state error is a constant. 10. Why does system type depend on where the external signal enters the system? Because the error depends on where the input enters, so does the type. 11. Give two reasons to use a digital controller rather than an analog controller. (a) The control law is easier to change if the controller is digital. (b) A digital controller can perform logic and other nonlinear operations much easier than an analog controller. (c) The hardware of a digital controller can be fixed in the design before the details of the actual control design are finished. 12. Give two disadvantages to using a digital controller. (a) The bandwidth of a digital controller is limited by the possible sample frequency. (b) The digital controller introduces noise by the quantization process. 13. Give the substitution in the discrete operator z for the Laplace operator s if the approximation to the integral in Eq. (4.98) is taken to be the rectangle of height e(kTs ) and base Ts . s=

z−1 Ts

Chapter 5 1. Give two definitions for the root locus. (a) The root locus is the locus of points in the s-plane where the equation a(s) + Kb(s) = 0 has a solution.

778 Appendix B Solutions to the Review Questions (b) The root locus is the locus of points in the s-plane where the angle of G(s) = b(s)/a(s) is 180◦ . 2. Define the negative root locus. The negative root locus is the locus of points where the equation a(s)−Kb(s) = 0 has a solution or where the angle of G(s) = b(s)/a(s) is 0◦ . 3. Where are the sections of the (positive) root locus on the real axis? Segments of the real axis to the left of an odd number of zeros and poles are on the root locus. 4. What are the angles of departure from two coincident poles at s = −a on the real axis? Assume there are no poles or zeros to the right of −a. The loci depart at ±90◦ . 5. What are the angles of departure from three coincident poles at s = −a on the real axis? Assume there are no poles or zeros to the right of −a. The loci depart at ±60◦ and 180◦ . 6. What is the requirement on the location of a lead compensation zero that will cause the locus to pass through a desired root location, ro ? The angle from the lead zero to ro must be such that the angle of the compensated transfer function at ro is 180◦ . 7. What is the value of the compensator gain that will cause a closed-loop pole to be at ro ? If the compensated open-loop transfer function is KDG, then K = 1/ |D(ro )G(ro )| . 8. What is the principal effect of a lead compensation on a root locus? The lead compensation generally causes the locus to bend toward the LHP, moving the dominant roots to a place of higher damping. 9. What is the principal effect of a lag compensation on a root locus in the vicinity of the dominant closed-loop roots? The lag compensation is normally placed so near the origin that it has negligible effect on the root locus in the vicinity of the dominant closed-loop roots. 10. What is the principal effect of a lag compensation on the steady-state error to a polynomial reference input? A lag compensation normally raises the gain at s = 0 and thus increases the velocity constant of a Type 1 system and lowers the error to polynomial inputs. 11. Why is the angle of departure from a pole near the imaginary axis especially important? If the locus starts toward the RHP, then feedback will make the system less stable. On the other hand if the locus departs toward the LHP, then feedback is going to make the system more stable. 12. Define a conditionally stable system.

Appendix B Solutions to the Review Questions

779

A system that becomes unstable as gain is reduced is considered to be conditionally stable. That is, its stability is conditioned on having an operating compensator with at least a minimum value of gain. 13. Show, with a root locus argument, that a system having three poles at the origin must be conditionally stable. With three poles at the origin, the angles of departure ensure that two poles leave the origin at 180◦ , ±60◦ , or, if there are poles on the real axis in the RHP, they may leave at 0◦ , ±120◦ which is to say that at least one pole begins by moving into the RHP. As gain is reduced from the operating level, at least one root must pass into the RHP for gain low enough; therefore, the system must be conditionally stable.

Chapter 6 1. Why did Bode suggest plotting the magnitude of a frequency response on log– log coordinates? In log–log coordinates, the plot for a rational transfer function can be well guided by linear asymptotes and thus easily plotted and visualized. 2. Define a decibel. If a power ratio is P1 /P2 , then the measure in decibels is 10 log(P1 /P2 ). Because power is proportional to voltage squared, and a transfer function would give a ratio of voltages, then the gain of a transfer function G(jω) in decibels is Gdb = 20 log |G(jω)|. 3. What is the transfer function magnitude if the gain is listed as 14 db? 14 = 20 log M, therefore M = 5.01. 4. Define gain crossover. The gain crossover ωc is the value of frequency where the magnitude gain is 1 (or 0 db). 5. Define phase crossover. The phase crossover ωcp is the value of the frequency where the phase crosses −180◦ . 6. Define phase margin, PM. The phase margin PM is a measure of how far in phase the Nyquist plot is from instability. In the typical case, if the phase of the system at gain crossover is φ, then the phase margin is 180◦ + φ. For example, if φ = −150◦ , then the phase margin is 30◦ . 7. Define gain margin, GM. The gain margin is a measure of how far the system is from instability by changes ◦ in If the gain at phase crossover, where    the system phase is 180 , is   gain alone. G(jωcp ), then the gain margin is GM∗ G(jωcp ) = 1.0 or GM = 1/G(jωcp ). 8. What Bode plot characteristic is the best indicator of the closed-loop step response overshoot?

780 Appendix B Solutions to the Review Questions The phase margin is related to the equivalent closed-loop damping ratio approximately by ζeq = PM/100. As we saw in Chapter 3, the step response overshoot is monotonically related to the damping ratio. 9. What Bode plot characteristic is the best indicator of the closed-loop step response rise time? The rise time is measured by the closed-loop natural frequency, which in turn is adequately approximated by the gain crossover. Thus the best indicator of rise time is ωcg . 10. What is the principal effect of a lead compensation on Bode plot performance measures? The lead compensation usually is used to raise the phase margin at a desired gain crossover frequency. 11. What is the principal effect of a lag compensation on Bode plot performance measures? The lag compensation is usually used to raise the low-frequency gain to reduce the steady-state error to polynomial or low-frequency sinusoidal inputs. It can also be used to lower the crossover frequency ωc , where a more favorable phase exists. 12. How do you find the Kv of a type 1 system from its Bode plot? The Kv is determined by the low-frequency asymptote, which has a slope of −1 for a type 1 system and is given by the expression Kv /ω. The value of the constant may be found either from the frequency where the asymptote reaches 1.0 (or 0 db) or else as the value of the asymptote at the frequency of ω = 1. 13. Why do we need to know beforehand the number of open-loop unstable poles in order to tell stability from the Nyquist plot? The Nyquist plot encirclements counts the difference in the number of zeros and the number of poles in the RHP of 1 + KDG. In order to know the number of zeros of this function (which are closed-loop poles and thus unstable poles of the closed loop), we must know the number of unstable open-loop poles for the plot. 14. What is the main advantage in control design of counting the encirclements of −1/K of D(jω)G(jω) rather than encirclements of −1 of KD(jω)G(jω)? If we plot DG alone, then the stability depends on the encirclements of −1/K. The designer can thus easily look at the entire range of real K and determine the best value of gain for the design without having to make any more plots. 15. Define a conditionally stable feedback system. How can you identify one on a Bode plot? A conditionally stable system becomes unstable as gain is reduced. If the lowfrequency phase drops below −180◦ then a reduction in gain until gain crossover occurs where there is no phase margin, then the system is almost surely unstable. A look at the Nyquist plot is necessary to be certain. This condition can also be

Appendix B Solutions to the Review Questions

781

seen easily from a root locus; the locus will have segments in the RHP for low values of gain. 

16. A certain control system is required to follow sinusoids, which may be any frequency in the range 0 ≤ ω ≤ 450 rad/sec and have amplitudes up to 5 units with (sinusoidal) steady-state error to be never more than 0.01. Sketch (or describe) the corresponding performance function W1 (ω). The magnitude of W1 is given by the ratio |R| /eb = 5/0.01 = 500. The performance function would then have the value 500 for frequencies up to 450 rad/sec. The Bode magnitude plot would be required to be above this curve for these frequencies.

Chapter 7 The following questions are based on a system in state-variable form with matrices F, G, H, and J, input u, output y, and state x: 1. Why is it convenient to write equations of motion in state-variable form? It provides a standard way to describe the differential equations for any dynamic system so that computer-aided analysis can be carried out more conveniently. It is also more convenient to analyze linear systems in terms of the standard description matrices. 2. Give an expression for the transfer function of this system. G(s) = H(sI − F)−1 G + J. 3. Give two expressions for the poles of the transfer function of the system. (a) p = eig(F). (b) p = roots of det[sI − F] = a(s) = 0. 4. Give an expression for the zeros of the system transfer function.   sI − F −G z = roots of det = b(s) = 0. H J 5. Under what condition will the state of the system be controllable? (a) If the pair (F, G) is controllable—that is, if the matrix  C = G FG · · · Fn−1 is full rank. (b) If the system can be put into control canonical form. 6. Under what conditions will the system be observable from the output y?

782 Appendix B Solutions to the Review Questions (a) If the matrices (F, H) are observable—that is, if the matrix ⎡

H ⎢ HF ⎢ O=⎢ .. ⎣ . HF(n−1)

⎤ ⎥ ⎥ ⎥ ⎦

has full rank. (b) If the system can be put into observable canonical form. 7. Give an expression for the closed-loop poles if state feedback of the form u = −Kx is used. (a) pc = eig( F − G ∗ K). (b) pc = roots of det(sI − F + GK) = αc (s) = 0. 8. Under what conditions can the feedback matrix K be selected so that the roots of αc (s) are placed arbitrarily? If the system is controllable. 9. What is the advantage of using the LQR or symmetrical root locus in designing the feedback matrix K? With LQR, the closed-loop system will be more robust to parameter changes, and the designer has some control over the control effort used by the closed-loop system. 10. What is the main reason for using an estimator in feedback control? When the state is not available (usually because it is too expensive or impractical to put sensors on each state variable), then an estimator using only the output y can give an estimate that can be used in place of the actual state. 11. If the estimator gain L is used, give an expression for the closed-loop poles due to the estimator. (a) pe = eig(F − L ∗ H). (b) pe = roots of det(sI − F + LH) = αe (s) = 0. 12. Under what conditions can the estimator gain L be selected so that the roots of αe (s) = 0 are placed arbitrarily? If the system is observable. 13. If the reference input is arranged so that the input to the estimator is identical to the input to the process, what will be the overall closed-loop transfer function? T (s) = N

b(s) . αc (s)

Appendix B Solutions to the Review Questions

783

14. If the reference input is introduced in such a way as to permit the zeros to be assigned as the roots of γ (s), what will the overall closed-loop transfer function be? γ (s)b(s) , T (s) = N αe (s)αc (s) usually γ (s) = αe (s). 15. What are the three standard techniques for introducing integral control in the state-feedback design method? (a) By augmenting the process state to include an integrator state variable. (b) By the internal model approach. (c) By using the extended estimator approach.

Chapter 8 1. What is the Nyquist rate? What are its characteristics? The Nyquist rate is half the sample rate, or = ωs /2. Above this rate, no frequencies can be represented by a sampled signal. 2. Describe the discrete equivalent design process. The controller for a system is designed as if the controller will be analog. The resulting controller is then approximated by a digital equivalent. 3. Describe how to arrive at a D(z) if the sample rate is 30 × ωBW . Use the discrete equivalent design method. It typically yields satisfactory results for such a high sample rate. But after using the discrete equivalent, check the result using a simulation that includes the effect of sampling or else perform an exact discrete linear analysis. It is best to use a simulation that includes all known sampling effects and system delays. 4. For a system with a 1 rad/sec bandwidth, describe the consequences of various sample rates. An absolute minimum sample rate is 2 rad/sec (or 0.32 Hz and T = 3 sec). From 2 rad/sec to 10 or 20 rad/sec, the control will be jerky with noticeable steps in the control and the design needs to be done very carefully. Between 20 and 30 rad/sec, the magnitude of the control steps become progressively smaller and design using discrete equivalents works reasonably well. Above 30 rad/sec, the control steps are hardly noticeable and the discrete equivalent can be used with confidence. 5. Give two advantages for selecting a digital processor rather than analog circuitry to implement a controller. (a) The physical layout of a digital controller can be done before the final design is complete, often resulting in completing the hardware implementation in

784 Appendix B Solutions to the Review Questions

(b)

(c)

(d)

(e)

much less time than required to get an analog controller specified and constructed. A digital processor is more flexible in making design changes as software is easier to reprogram than rewiring and/or adding op-amps to a printed circuit board. A digital processor can much more easily include nonlinear terms and logic decision steps in the overall controller design to permit adaptive control or gain scheduling, for example. Many models of the same basic controller can be accommodated by simply using different PROMS with the same hardware design. For example, an automobile manufacturer might have one engine controller hardware design for its entire product line; but have a different PROM for each engine/vehicle combination. Digital controllers are less sensitive to temperature variations than analog controllers.

6. Give two disadvantages of selecting a digital processor rather than analog circuitry to implement a controller. (a) The finite sampling rate of the A/D and D/A converters and the finite compute speed of the processor limit the bandwidth of the controller to about 1/10 of the sample frequency. (b) The finite accuracy or bit length of the converters introduce extra noise or offsets into the control loop if using low-end controllers. (c) Cost. For simple controllers, a digital implementation will typically be more expensive than an analog implementation. 

7. Describe how to arrive at a D(z) if the sample rate is 5 × ωBW . Start by using the discrete equivalent, but include an approximation of the effect of the delay in the plant model when carrying out the analog design. Then check the result via an exact discrete analysis by converting the plant to its discrete equivalent and combining that with the discrete controller. If performance is degraded from that desired, modify the discrete controller using discrete design methods. Finish by using a simulation that includes all known sampling effects and system delays.

Chapter 9 1. Why do we approximate a physical model of the plant (which is always nonlinear) with a linear model? Analysis and design of linear models is vastly simpler than with nonlinear models. Furthermore, it has been shown (by Lyapunov) that if the linear approximation is stable, then there is at least some region of stability for the nonlinear model.

Appendix B Solutions to the Review Questions

785

2. How would you linearize the nonlinear system equation for radiation heat transfer T˙ = T 4 + T + u? δ T˙ = (4To3 + 1)δT + δu, where To is the nominal operating temperature. (See the RTP case study in Chapter 10.) 3. A lamp used as a thermal actuator has a nonlinearity such that the experimentally measured output power is related to the input voltage by P = V 1.6 . How would you deal with such a nonlinearity in feedback control design? We precede the lamp with an inverse nonlinearity—that is, V = P0.625 —so as to linearize the cascaded system (see the RTP case study in Chapter 10). 4. What is integrator windup? If the plant actuator output signal saturates, then it may take a long time for the error to be brought back to zero from an initial upset and during this time the integrator output may grow or “windup” much more than it would if the system were linear. Special “antiwindup” circuits are designed to prevent windup. 5. Why is an antiwindup circuit important? When a control includes integral action and is subject to saturation, large inputs can cause large overshoots and slow recovering unless an antiwindup circuit is included. 6. Using the nonlinear saturation function having gain 1 and limits ±1, sketch the block diagram of saturation for an actuator that has gain 7 and limits ±20. If the output of the actuator is uout and itsinput  is uin , the control is given by 7uin uout = 20 sat . 20 7. What is a describing function and how is it related to a transfer function? The goal of the describing function approach is to find something like a “transfer function” for a nonlinear element. One may view the describing function as an extension of the frequency response to nonlinearities. 8. What are the assumptions behind the use of the describing function? The basic assumption is that the plant behaves approximately as a low-pass filter. The other assumptions are that the nonlinearity is time invariant, and there is a single nonlinear element in the system. 9. What is a limit cycle in a nonlinear system? In some nonlinear systems the error builds up and the response approaches a periodic solution of fixed amplitude, the limit cycle, as time grows large. 10. How can one determine the describing function for a nonlinear system in the laboratory? We can inject sinusoidal signals into the system and place a low-pass filter with a sharp cutoff at the output of the system to measure the fundamental component of the output. The describing function is then computed as the ratio of the amplitude of the fundamental component of the output of the nonlinear system over the amplitude of the sinusoidal input signal.

786 Appendix B Solutions to the Review Questions 11. What is the minimum time-control strategy for a satellite attitude control with bounded controls? Bang-bang. 12. How are the two Lyapunov methods used? His indirect or first method is based on linearization of the equations of motion and drawing conclusions about the stability of the nonlinear system by considering the stability of the linear approximation. In his direct or second method, the nonlinear equations are considered directly.

Chapter 10 1. Why is a collocated actuator and sensor arrangement for a lightly damped structure such as a robot arm easier to design than a noncollocated setup? In the collocated case, the process naturally has zeros near the lightly damped poles, which keep the root locus in the LHP. 2. Why should the control engineer be involved in the design of the process to be controlled? In many cases, the characteristics and locations of the actuators and sensors can have a major impact on the complexity and difficulty in design of the controller. If the needs of control are included in the process design, the final systems are often more effective (better closed-loop performance) and less expensive. 3. Give examples of an actuator and a sensor for the following control problems: (a) Attitude control of a geosynchronous communication satellite. Actuators: Cold gas-jet thrusters, momentum wheels, magnetic torquers (coils, torque rod), plasma thruster. Sensors: Earth sensor (roll, pitch), digital integrated rate assembly (DIRA) gyro (for rates), star tracker. (b) Pitch control of a Boeing 747 airliner. Actuators: Elevator. Sensors: Pitch rate and/or pitch angle is measured using a gyro or a ringlaser gyro. (c) Track-following control of a CD player. Actuators: DC motor to move the (dual stage sledge) arm mechanism, magnetic coils (two) for focusing on tracks. Sensors: Array of photodiodes. (d) Fuel–air ratio control of a spark-ignited automobile engine. Actuators: Fuel injection. Sensors: Zirconium oxide sensor. (e) Position control for an arm of a robot used to paint automobiles. Actuators: Hydraulic actuators or electric motors. Sensors: Encoders to measure arm rotations, pressure sensors, and force sensors.

Appendix B Solutions to the Review Questions

787

(f) Heading control of a ship. Actuators: Rudder. Sensors: Gyrocompass. (g) Attitude control of a helicopter. Actuators: Moving swash plate (either via direct link or servo) rotates main blade angle of attack. Sensors: Same as aircraft (pitot tube, accelerometers, rate gyros).

Appendix C MATLAB® Commands MATLAB function (.m file) or Variable Description angle ans abs acker atan2 axis bilin bode bodemag c2d canon clear clf close close all conj conv cos ctrb ctrbf damp dcgain deconv det diag diary dstep eig exp expm eye ezplot feedback

788

Page (s)

Phase Angle Most recent answer Absolute value Ackermann’s formula for pole placement 448,456,469,472,504 Four quadrant inverse tangent Control axis scaling 329,332 Bilinear transform Bode frequency response 84,300,315,380,522 Bode magnitude frequency response Continuous-to-discrete conversion 203,353,488,499,570 State-space canonical forms 435 Clear variables and functions Clear current figure Close figure Close all figures Complex conjugate Polynomial multiplication 92,459,461,767 Cosine Controllability matrix Staircase canonical form, controllability 430 Damping and natural frequency 588–589 Computes DC gain of LTI system Division of polynomials Determinant of a matrix Diagonal matrix, diagonals of a matrix Save text of MATLAB session Step response of a discrete system Eigenvalues and eigenvectors Exponential 432 Matrix exponential Identity matrix Easy-to-use function plotter Feedback connection of two systems 262–263

Appendix C MATLAB® Commands MATLAB function (.m file) or Variable Description figure figure(i) find format freqresp gram grid hold i ilaplace imag impulse inf initial inv j laplace linmod linmod2 line linspace load log log10 loglog logspace lqe lqr lsim ltiview ltru ltry margin max mean min nan nichols norm nyquist obsv obsvf ones

Page (s)

Create figure window Make i the current figure Find indices of nonzero elements Set output format Frequency response of LTI systems Controllability/observability Gramian Grid lines Hold √ current plot −1 Inverse Laplace transform Complex imaginary part Impulse response of LTI system 111,115,127 Infinity Initial condition response of state-space system 469,475 Matrix inverse 434,436 √ −1 Laplace transform 86 Linearization 620 Linearization (advanced) 603 Create a line Linearly spaced vector Load in workspace variables Natural logarithm Logarithm to the base 10 Log–log plot 84,300,315,522 Logarithmically spaced frequency points 84 Linear Quadratic Estimator design 521 Linear Quadratic Regulator design 463,464,725 Simulation of LTI system with arbitrary input 101,102 Opens the LTI viewer GUI Loop transfer recovery (LTR) Loop transfer recovery (LTR) Gain and phase margins 357,398,522 Largest component 380 Average or mean value Smallest component Not-a-number Nichol’s chart 383 Matrix or vector norm Nyquist plot 326,329,333 Observability matrix 471 Staircase canonical form, observability 471 Array of ones 101

789

790 Appendix C MATLAB® Commands MATLAB function (.m file) or Variable Description pade parallel place pi plot pole poly polyval printsys pzmap rand randn rank real residue rlocfind rlocus rltool roots save semilogx semilogy series sgrid sin sim sisotool size sort sqrt squeeze ss2ss ss2tf ss2zp ss ssdata std step subplot sum svd syms text

Page (s)

Pade approximation for time delay 273 Parallel connection of two LTI systems 107 Pole placement 451,472,484,490,510 3.141592653589793 Plot function 33,99,101 Poles of LTI system Form polynomial from its roots 96 Evaluate polynomial Print system in pretty format 98 Pole–zero map 109 Uniformly distributed random numbers Normally distributed random numbers Matrix rank Complex real part Residues in partial-fraction expansion 92,96,767 Find Root-Locus gain 246,399 Root locus 225,399,459,461 Interactive Root-Locus tool 282 Roots of a polynomial 136,239,441,451 Save workspace variables Semi-log plot 84,300,522 Semi-log plot 380 Series connection of two LTI systems 107,522,588–589 s-plane grid lines Sine Simulate a SIMULINK model SISO design tool Size of an array Sort in ascending or descending order Square root 510 Remove singleton dimensions State-space similarity transformation State space to transfer function conversion 438,440,441 State space to pole–zero conversion 423 Conversion to state-space 419,435,440,442 Create a state-space model 435 Standard deviation Step response 24,137,138 Multiple plots on the same window Sum of Elements Singular-value decomposition Declaration of symbolic variables 759 Text annotation

Appendix C MATLAB® Commands MATLAB function (.m file) or Variable Description tf2ss tf2zp tf tfdata title tzero var who why whos xlabel xlsread ylabel zero zeros zgrid zpk zp2tf

Page (s)

Transfer function to state-space conversion 426 Transfer function to pole–zero conversion 98,100 Creation or conversion to transfer function 24,84,107,111,127,201,459 Transfer function data Plot title Transmission zeros 468,479 Variance List of current variables Answers any question you may have List of current variables, long form x-axis label Get data from an Excel spreadsheet y-axis label Transmission zeros Array of zeros 101 z-plane grid lines Zero–pole–gain Zero–pole to transfer function conversion 100

791

This page intentionally left blank

Bibliography Abramovitch, D. and G. F. Franklin, “A brief history of disk drive control,” IEEE Control System Magazine, Vol. 22, pp. 28–42, June 2002. Ackermann, J., “Der entwurf linearer regelungssysteme im zustandsraum,” Regelungstech. Prozess-Datenverarb., Vol. 7, pp. 297–300, 1972. Airy, G. B., “On the regulator of the clock-work for effecting uniform movement of equatorials,” Mem. R. Astron. Soc., Vol. 11, pp. 249–267, 1840. Alon, U., An Introduction to Systems Biology. Chapman & Hall/CRC, 2007. Alon, U., M. G. Surette, N. Barkai, and S. Leibler, “Robustness in bacterial chemotaxis,” Nature, Vol. 397, pp. 168–171, January 1999. Anderson, B. D. O. and J. B. Moore, Optimal Control: Linear Quadratic Methods. Upper Saddle River, NJ: Prentice Hall, 1990. Anderson, E., et al., LAPACK User’s Guide, 3rd ed., Philadelphia, PA: SIAM, 1999. Åström, K. J., “Frequency domain properties of Otto Smith regulators,” Int. J. Control, Vol. 26, No. 2, pp. 307–314, 1977. Åström, K. J. and T. Hägglund, PID Controllers: Theory, Design, and Tuning, 2nd ed., Research Triangle, NC: International Society for Measurement and Control, 1995. Åström, K. J. and T. Hägglund, Advanced PID Control. Research Triangle, NC: International Society for Measurement and Control, 2006. Athans, M., “A tutorial on the LQG/LTR method,” Proc. American Control Conf, pp. 1289–1296, June 1986. Barkai, N. and S. Leibler, “Robustness in simple biochemical networks,” Nature, Vol. 387, pp. 913–917, 1997. Bellman, R. and R. Kalaba, eds., Mathematical Trends in Control Theory. New York: Dover, 1964. Berg, H. C., E. coli in Motion. New York: Springer-Verlag, 2004. Bergen, A. R. and R. L. Franks, “Justification of the describing function method,” SIAM J. Control, Vol. 9, pp. 568–589, 1971. Blakelock, J. H., Automatic Control of Aircraft and Missiles. 2nd ed., NewYork: John Wiley, 1991. Bodanis, D., E = MC 2 : A Biography of the World’s Most Famous Equation. New York: Walker and Co., 2000. Bode, H. W., “Feedback: The history of an idea,” Conference on Circuits and Systems. New York (1960): Reprinted in Bellman and Kalaba, 1964. —– Network Analysis and Feedback Amplifier Design. New York: Van Nostrand, 1945. Boyd, S. P. and C. H. Barratt, Linear Controller Design: Limits of Performance. Upper Saddle River, NJ: Prentice Hall, 1991.

793

794 Bibliography Brennan, R. P., Heisenberg Probably Slept Here. New York: John Wiley & Sons, 1997. Brown, J. W. and R. V. Churchill, Complex Variables and Applications. 6th ed., New York: McGraw-Hill, 1996. Bryson, A. E., Jr. and W. F. Denham, “A steepest-ascent method for solving optimum programming problems,” J. Appl. Mech., June 1962. Bryson, A. E., Jr. and Y. C. Ho, Applied Optimal Control. Waltham, MA: Blaisdell, 1969. Bryson, A. E., Jr., Control of Spacecraft and Aircraft. Princeton, NJ: Princeton University Press, 1994. Callender, A., D. R. Hartree, and A. Porter, “Time lag in a control system,” Philos. Trans. R. Soc. London A, London: Cambridge University Press, 1936. Campbell, G. A. and R. N. Foster, Fourier Integrals for Practical Applications. New York: Van Nostrand, 1948. Campbell, N. A. and J. B. Reece, Biology, 8th ed., Benjamin Cummings, 2008. Cannon, R. H., Jr., Dynamics of Physical Systems. New York: McGraw-Hill, 1967. Churchill, R. V., Operational Mathematics, 3rd ed., New York: McGraw-Hill, 1972. Clark, R. N., Introduction to Automatic Control Systems. New York: John Wiley, 1962. Clegg, J. C., “A nonlinear integrator for servomechanisms,” Trans. AIEE, Pt. II, Vol. 77, pp. 41–42, 1958. de Roover, D., L. Porter, A. Emami-Naeini, J. A. Marohn, S. Kuehn, S. Garner, and D. Smith, “An all-digital cantilever controller for MRFM and scanned probe microscopy using a combined DSP/FPGA design,” Am. Lab., Vol. 40, No. 8, pp. 12–17, 2008. de Roover, D., A. Emami-Naeini, and J. L. Ebert, “Model-based control of fast-ramp RTP systems,” Sixth International Conference on Advanced Thermal Processing of Semiconductors, pp. 177–186, Kyoto, Japan, September 1998. de Vries, G., T. Hillen, M. Lewis, J. Muller, and B. Schonfisch, A Course in Mathematical Biology. SIAM, 2006. Dorato, P., Analytic Feedback System Design: An Interpolation Approach, Pacific Grove, CA: Brooks/Cole, 2000. Doyle, J. C., B. A. Francis, and A. Tannenbaum, Feedback Control Theory. NewYork: Macmillan, 1992. Doyle, J. C. and G. Stein, “Multivariable feedback design: Concepts for a classical/ modern synthesis,” IEEE Trans. Autom. Control, Vol. AC-26, No. 1, pp. 4–16, February 1981. Doyle, J. C., “Guaranteed margins for LQG regulators,” IEEE Trans. Autom. Control, Vol. AC-23, pp. 756–757, 1978. Doyle, J. C. and G. Stein, “Robustness with observers,” IEEE Trans. Autom. Control, Vol. AC-24, pp. 607–611, August 1979.

Bibliography

795

Ebert, J. L., A. Emami-Naeini, H. Aling, and R. L. Kosut, “Thermal modeling of rapid thermal processing systems,” Third International Rapid Thermal Processing Conference, pp. 343–355, Amsterdam, August 1995a. Ebert, J. L., A. Emami-Naeini, and R. L. Kosut, “Thermal Modeling and control of rapid thermal processing systems,” Proceeding 34th IEEE Conference Decision and Control, pp. 1304–1309, December 1995b. Elgerd, O. I., Electric Energy Systems Theory. New York: McGraw-Hill, 1982. —– and W. C. Stephens, “Effect of closed-loop transfer function pole and zero locations on the transient response of linear control systems,” Trans. Am. Inst. Electr. Eng. Part 1, Vol. 42, pp. 121–127, 1959. Emami-Naeini, A., “The shapes of Nyquist plots: Connections with classical plane curves,” IEEE Control Systems Magazine, October 2009. Emami-Naeini, A., J. L. Ebert, D. de Roover, R. L. Kosut, M. Dettori, L. Porter, and S. Ghosal, “Modeling and control of distributed thermal systems,” Proc. IEEE Trans. Control Systems Technol., Vol. 11, No. 5, pp. 668–683, September 2003. Emami-Naeini, A. and G. F. Franklin, “Zero assignment in the multivariable robust servomechanism,” Proc. IEEE Conf. Dec. Control, pp. 891–893, December, 1982. Emami-Naeini, A., and P. Van Dooren, “Computation of zeros of linear multivariable systems,” Automatica, Vol. 18, No. 4, pp. 415–430, 1982. —– “On computation of transmission zeros and transfer functions,” Proc. IEEE Conf. Dec. Control, pp. 51–55, December 1982. Etkin, B. and L. D. Reid, Dynamics of Flight: Stability and Control, 3rd ed., NewYork: John Wiley, 1996. Evans, G. W., “Bringing root locus to the classroom,” IEEE Control Systems Magazine, Vol. 24, pp. 74–81, 2004. Freudenberg, J. S. and D. P. Looze, “Right half plane zeros and design tradeoffs in feedback systems,” IEEE Trans. Autom. Control, Vol. AC-30, pp. 555–561, June 1985. Fuller, A. T., “The Early development of control theory,” J. Dyn. Syst. Meas. Control, Vol. 98, pp. 109–118 and 224–235, 1976. Gardner, M. F. and J. L. Barnes, Transients in Linear Systems. New York: John Wiley, 1942. Gunckel, T. L., III and G. F. Franklin, “A general solution for linear sampled data control,” J. Basic Eng., Vol. 85-D, pp. 197–201, 1963. Gyugyi, P., Y. Cho, G. F. Franklin, and T. Kailath, “Control of rapid thermal processing: A system theoretic approach,” Proceedings of IFAC World Congress, 1993. Hanselman, D. C. and B. C. Littlefield, Mastering MATLAB 7, Upper Saddle River. NJ: Prentice Hall, 2005. Heffley, R. K. and W. F. Jewell, Aircraft Handling Qualities, Technical Report 1004-1, System Technology, Inc., Hawthorne, CA, May 1972.

796 Bibliography Higham, D. J. and N. J. Higham, MATLAB Guide, 2nd ed., Philadelphia: SIAM, 2005. Ho, M.-T., A. Datta, and S. P. Bhattacharyya, “An elementary derivation of the RouthHurwitz criterion,” IEEE Trans. Autom. Control, Vol. 43, No. 3, pp. 405–409, 1998. Huang, J-J. and D. B. DeBra, “Automatic tuning of Smith-predictor design using optimal parameter mismatch,” Proc. IEEE Conf. Dec. Contr., pp. 3307–3312, December 2000. Hubbard, M., Jr., and J. D. Powell, “Closed-loop control of internal combustion engine exhaust emissions,” SUDAAR No. 473, Department of Aero/Astro, Stanford University, Stanford, CA, February 1974. James, H. M., N. B. Nichols, and R. S. Phillips, Theory of Servomechanisms, Radiation Lab. Series, 25. New York: McGraw-Hill, 1947. Johnson, R. C., Jr., A. S. Foss, G. F. Franklin, R. V. Monopoli, and G. Stein, “Toward development of a practical benchmark example for adaptive control,” IEEE Control System Magazine, Vol. 1, No. 4, pp. 25–28, December 1981. Joseph, P. D. and J. T. Tou., “On linear control theory,” AIEE Transactions, Vol. 80, pp. 193–196, 1961. Kailath, T., Linear Systems. Upper Saddle River, NJ: Prentice Hall, 1980. Kalman, R. E., “A new approach to linear filtering and prediction problems,” J. Basic Eng., Vol. 85, pp. 34–45, 1960a. Kalman, R. E. and J. E. Bertram, “Control system analysis and design via the second method of Lyapunov. II. Discrete Systems,” J. Basic Eng., Vol. 82, pp. 394–400, 1960. Kalman, R. E., Y. C. Ho, and K. S. Narendra, “Controllability of linear dynamical systems,” Contributions to Differential Equations, Vol. 1. New York: John Wiley, 1962. Khalil, H. K., Nonlinear Systems, 3rd ed., Upper Saddle River, NJ: Prentice Hall, 2002. Kharitonov, V. L., “Asymptotic stability of an equilibrium position of a family of systems of linear differential equations,” Differential’nye Uraveniya, Vol. 14, pp. 1483–1485, 1978. Kochenburger, R. J., “A frequency response method for analyzing and synthesizing contactor servomechanisms,” Trans. Am. Inst. Electr. Eng., Vol. 69, pp. 270–283, 1950. Kuo, B. C., ed., Incremental Motion Control, Vol. 2: Step Motors and Control Systems, Champaign, IL: SRL Publishing, 1980. —– Proceedings of the Symposium Incremental Motion Control Systems and Devices, Part. 1: Step Motors and Controls, Champaign-Urbana, IL: University of Illinois, 1972. Lanchester, F. W., Aerodonetics. London: Archibald Constable, 1908. LaSalle, L. P. and S. Lefschetz, Stability by Lyapunov’s Direct Method. New York: Academic Press, 1961.

Bibliography

797

Lyapunov, A. M., “Problème général de la stabilité du mouvement,” Ann. Fac. Sci. Univ. Toulouse Sci. Math. Sci. Phys., Vol. 9, pp. 203–474, 1907; original paper published in 1892 in Commun. Soc. Math. Kharkow, 1892; reprinted as Vol. 17 in Annals of Math Studies. Princeton, NJ: Princeton University Press, 1949. Ljüng, L., System Identification: Theory for the User, 2nd ed., Upper Saddle River, NJ: Prentice Hall, 1999. Luenberger, D. G., “Observing the state of a linear system,” IEEE Trans. Mil. Electron., Vol. MIL-8, pp. 74–80, 1964. Mahon, B., The Man who Changed Everything: The Life of James Clerk Maxwell, UK: Wiley, 2003. Marsden, J. E. and M. J. Hoffman, Basic Complex Analysis, 3rd ed., Freeman, 1999. Mason, S. J., “Feedback theory: Some properties of signal flow graphs,” Proc. IRE, Vol. 41, pp. 1144–1156, 1953. —– “Feedback theory: Further properties of signal flow graphs,” Proc. IRE, Vol. 44, pp. 920–926, 1956. Maxwell, J. C., “On governors,” Proc. R. Soc. Lond., Vol. 16, pp. 270–283, 1868. Mayr, O., The Origins of Feedback Control. Cambridge, MA: MIT Press, 1970. McRuer, D. T., I. Askenas, and D. Graham, Aircraft Dynamics and Automatic Control, Princeton, NJ: Princeton University Press, 1973. Mello, B. A., L. Shaw, and Y. Tu, “Effects of receptor interaction in bacterial chemotaxis,” Biophysical Journal, Vol. 87, pp. 1578–1595, September 2004. Messner, W. C. and D. M. Tilburry, Control Tutorials for MATLAB and Simulink: A Web-Based Approach, Upper Saddle River, NJ: Prentice Hall, 1999. Minimis, G. S. and C. C. Paige, “An algorithm for pole assignment of time invariant systems,” Int. J. Control, Vol. 35, No. 2, pp. 341–354, 1982. Moler, C. B., “Nineteen dubious ways to compute the exponential of a matrix, twentyfive years later,” SIAM Rev., Vol. 45, No. 1, pp. 3–49, 2003. —– Numerical Computing with MATLAB, Philadelphia, PA: SIAM, 2004. Norman, S. A., “Wafer temperature control in rapid thermal processing,” Ph.D. Dissertation, Stanford University, Stanford, CA, 1992. Nyquist, H., “Regeneration theory,” Bell Systmes Technical. J., Vol. 11, pp. 126–147, 1932. Oswald, R. K., “Design of a disk file head positioning servo,” IBM J. Res. Dev., Vol. 18, pp. 506–512, November 1974. Parks, P., “Lyapunov redesign of model reference adaptive control systems,” IEEE Trans. Autom. Control, AC-11, No. 3, 1966. Perkins, W. R., P. V. Kokotovic, T. Boureret, and J. L. Schiano, “Sensitivity function methods in control system education,” IFAC Conference on Control Education, June 1991. Ragazzini, J. R., R. H. Randall, and F. A. Russell, “Analysis of problems in dynamics by electronic circuits,” Proc. IRE, Vol. 35, No. 5, pp. 442–452, May 1947.

798 Bibliography Ragazzini, J. R. and G. F. Franklin, Sampled-Data Control Systems, New York: McGraw-Hill, 1958. Reliance Motion Control Corp., DC Motor Speed Controls Servo Systems, 5th ed., Eden Prairie, MN: Reliance Motion Control Corp., 1980. Routh, E. J., Dynamics of a System of Rigid Bodies. London: MacMillan, 1905. Saberi, A., B. M. Chen, P. Sannuti, Loop Transfer Recovery: Analysis and Design. New York: Springer-Verlag, 1993. Safonov, M. G. and G. Wyetzner, “Computer-aided stability analysis renders Popov criterion obsolete,” IEEE Trans. Autom. Control, Vol. AC-32, pp. 1128–1131, 1987. Sandberg, I. W., “A frequency domain condition for stability of feedback systems containing a single time varying nonlinear element,” Bell Systems Technical J., Vol. 43, pp. 1581–1599, 1964. Sastry, S. S., Nonlinear Systems; Analysis, Stability, and Control. New York: Springer-Verlag, 1999. Schmitz, E., “Robotic arm control,” Ph.D. Dissertation, Stanford University, Stanford, CA, 1985. Sedra, A. S., and K. C. Smith, Microelectronics Circuits, 3rd ed., New York: Oxford University Press, 1991. Simon, H. A., “Dynamic programming under uncertainty with a quadratic function,”Econometrica, Vol. 24, pp. 74–81, 1956. Sinha, N. K. and B. Kuszta, Modeling and Identification of Dynamic Systems. New York: Van Nostrand, 1983. Smith, O. J. M., Feedback Control Systems. New York: McGraw-Hill, 1958. Sobel, D., Galileo’s Daughter. New York: Penguin Books, 2000. Stein, G. and M.Athans, “The LQG/LTR procedure for multivariable feedback control design,” IEEE Trans. Autom. Control, Vol. AC-32, pp. 105–114, February 1987. Strang, G., Linear Algebra and Its Applications, 3rd ed., New York: Harcourt Brace, 1988. Swift, J., On Poetry: A Rhapsody, 1973, J. Bartlett, ed., Familiar Quotations, 15th ed., Boston: Little Brown, 1980. Sze, S. M., ed. VLSI Technology, 2nd ed., New York: McGraw-Hill, 1988. Taubman, P., Secret Empire: Eisenhower, the CIA, and the Hidden Story of America’s Space Espionage, New York: Simon and Schuster, 2003. Thomson, W. T. and M. D. Dahleh, Theory of Vibration with Applications, 5th ed., Upper Saddle River, NJ: Prentice Hall, 1998. Trankle, T. L., “Development of WMEC Tampa maneuvering model from sea trial data,” Report MA-RD-760-87201. Palo Alto, CA: Systems Control Technology, March 1987. Truxal, J. G., Control System Synthesis. New York: McGraw-Hill, 1955.

Bibliography

799

van der Linden, G., J. L. Ebert, A. Emami-Naeini, and R. L. Kosut “RTP robust control design: Part II: controller synthesis,” Fourth International Rapid Thermal Processing Conference, pp. 263–271, September 1996. Van der Pol, B., and H. Bremmer, Operational Calculus. New York: Cambridge University Press, 1955. Van Dooren, P., A. Emami-Naeini, and L. Silverman, “Stable extraction of the kronecker structure of Pencils,” Proc. IEEE Conf. Dec. Control, San Diego, CA, pp. 521–524, December 1978. Vidyasagar, M., Nonlinear Systems Analysis, 2nd ed., Upper Saddle River, NJ: Prentice Hall, 1993. Wiener, N., “Generalized harmonic analysis,” Acta Math., Vol. 55, pp. 117, 1930. Woodson, H. H. and J. R. Melcher, Electromechanical Dynamics, Part I: Discrete Systems. New York: John Wiley, 1968. Workman, M. L., “Adaptive proximate time-optimal servomechanisms,” Ph.D. Dissertation, Stanford University, Stanford, CA, 1987. Yi, T.-M., Y. Huang, M. I. Simon, and J. C. Doyle, “Robust perfect adaptation in bacterial chemotaxis through integral feedback control,” PNAS, Vol. 97, No. 9, pp. 4649–4653, April 2000. Zames, G., “On the input-output stability of time-varying nonlinear feedback systems—Part I: Conditions derived using concepts of loop gain, conicity and positivity,” IEEE Trans. Autom. Control, Vol. AC-11, pp. 465–476, 1966. —– “On the input-output stability of time-varying nonlinear feedback systems—Part II: Conditions involving circles in the frequency plane and sector nonlinearities,” IEEE Trans. Autom. Control, Vol. AC-11, pp. 228–238, 1966. Ziegler, J. G. and N. B. Nichols, “Optimum settings for automatic controllers,” Trans. ASME, Vol. 64, pp. 759–768, 1942. —– “Process lags in automatic control circuits,” Trans. ASME, Vol. 65, No. 5, pp. 433–444, July 1943.

This page intentionally left blank

Index A Abscissa of convergence, 757 AC motor actuators, 49 acker.m, 449, 456, 469 Ackermann’s estimator formula, 471 pole placement, 448 SRL, 458 undamped oscillator, example of, 448 Activity, E. coli, 735, 736 Actuator saturation, feedback system with, 616 Actuators, 4 control system design procedure, 663 sizing, 619 Adams prize, 12 Adaptation, 735, 736 Adaptive control, 640 A/D converters, 577 digitization, 559 Additional poles effect of, 120–128 locus branches, 243 Agilent Technologies, 301 Aircraft lateral and longitudinal control control system design procedure, 684 nonlinear equations, 684–688 response using MATLAB, 127–128 Aircraft coordinates, 685 Airy, G. B., 11, 204 Aizerman’s conjecture, 644 Aliasing example, 579 z and s-plane, 566 Altimeter, 696

Altitude-hold autopilot, 696–702 Amplitude, armature voltage, 49 Amplitude ratio, 83 Amplitude scaling, 147 Amplitude stabilization, 371 Analog computer components, 421 implementation, 422–424 Analog implementation, 257–260 Analog prefilters, 578 Analog to digital (A/D) converters, 559, 577–578 Analysis tools, 583–585 Analytic continuation, 757 Anti-alias prefilters, 578–579 effect of, 582–583 Anti-windup compensation for PI controller, 618–619 Anti-windup methods, 616–617 Apollo, 740 Argument principle, 320–321 Armature, 48 Armature voltage, amplitudes of, 49 Arrival angles, rule for, 232 Assigned zeros, 494 Åström, Karl, 18, 557, 590 Asymptotes angles of, 229 center of positive locus, 230 frequency response, 308 positive locus, 229 Asymptotically stable, Lyapunov, 636 Asynchronous sampling, 583 Athans, M., 531 Attitude hold, 699 Attractant, 733

Augmented state equations with integral control, 503 Automobile suspension, 25 Automotive engine, control of fuel-air ratio in, 702–709 Autonomous estimator description, 493 example, 493, 494 Autopilot design block diagram, 262 root locus, 263 time response plots, 264 Auxiliary variable, 446

B Back emf laws and equations, 47 voltage, 47–49 Bacteria, 731 Ball levitator linearization of motion, 604–607 scaling, 147–148 Band center, 112 Band reject filter, 673 Bandwidth, 302 Bang-bang control, 632 Barkai, N., 734 Barometric altimeter, 696 Battin, Dick, 591, 740 Bell, Alexander Graham, 305 Bell Laboratories, 305 Bellman, R., 14 Berg, H., 733 Bertram, J. E., 640, 649 BIBO stability, 130 Bilateral Laplace transforms, 757 Bilinear approximation, 569 Binary fission, 732 Biological systems, 741

801

802 Index Black, H. S., 12, 205, 386 Block diagram, 6, 102–107 algebra, 105 reduction using MATLAB, 107–108 simplification example, 106 state equations, 425–436 third order system, 423 transfer function, 105–107 Blocking zeroes, 512, 514 Bode, H. W., 13, 205, 386 Bode gain phase relationship, 341–346 crossover frequency, 343 exercise problems related to, 401, 402 Bode gain phase theorem stated, 342 weighting function, graphically illustrated, 343 bode.m, 84, 300, 315 Bode plot complex poles, 311–313 complex poles and zeros, 313–315 computer-aided, 315 frequency response, advantages, 305 minimum and nonminimum phase systems, 316 multiple crossover frequency system, 340–341 real poles and zeros, 310–311 RHP, 332 rules, 309 Bode plot technique, frequency response, 304 Boeing 747 aircraft coordinates, 685 lateral and longitudinal control of, 684–702 Bounded input-bounded output (BIBO) stability, 130–131 Boyd, S.P., 523 Brahe, Tycho, 61 Break point, 307 Breakaway points, 224

Break-in point, 225 Bremmer, H., 87 Bridged tee circuit, 42–44 The Bristol company, 204 Bryson, A. E., 531 Bryson’s rule, 463 Butterworth configuration, 520

C c2d.m, 488 Campbell, G. A., 87 Cancellations, transfer function, 97 Canon, R. H., Jr., 52 Canonical forms block diagrams and, 425–427 control, 425 modal, 427 observer, 470 state equations, 425–436 canon.m, 435 Capacitor frequency response characteristics, 299 stability, 131 symbol and equation, 42 Carburetor, 702 Cart-stick balancer, 37 Cascaded tanks, 552 Case studies Automotive fuel/air ratio, 702 Boeing 747 lateral and longitude, 684 E. coli chemotaxis, 731 Hard-disk read/write head, 709 Satellite attitude, 668 Semiconductor wafer RTP, 717 Catalytic converter, 702 Cauchy’s Principle of the Argument, 320 Cell, 731 Characteristic equation, 439 closed-loop system, 444

Cheap control, 464 Chemotaxis, 733 Chemotaxis dynamics, 734 Chemotaxis model, 736 Circle criterion, 642, 644, 647 relationship with describing functions, 648 sector conditions, 643 sector for signum nonlinearity, 643 stability determination using, 648 Clark, R. N., 290 Classical control, 14 design, 415 Clegg integrator, 656 Closed loop bandwidth phase margin, 347 sample rate selection, 581 Closed loop control, 8, 170 Closed-loop estimator, 468 Closed loop frequency response, 346 exercise problems related to, 402 robust servomechanism, 513 Closed loop poles root locus, 271 Truxal’s formula, 495 Closed loop stability, frequency response determination, 318 Closed-loop system, 170, 172 characteristic equation, 444 LTR, 523 transfer function, 500 Closed loop transfer function, 302, 321–322 Closed-loop zeros, 495 Communication satellite, 28 Companion form matrix, 447 Comparator, 5 Compensation control law and estimator, 443 control system design procedure, 664 exercise problems related to, 402–409

Index frequency response designed method, 347–372 root locus, 248, 614 Compensator design (state-space), 549 conditionally stable, 485 exercise problems related to, 549, 554 full order, 480–482 reduced order, 482–483 transfer function, 521 Complementary sensitivity function, 376 Complete transient response, 85 Complex poles Bode plot, 311–315 s-plane plot, 113, 115 Complex zeros, Bode plot, 313–315 Composite curve, 309 Composite plot, 305 Computed torque, 600, 603 Computer, digital control, 579–580 Computer aids, 15 Computer-aided Bode plot for complex poles and zeros, 315 Conditionally stable compensator design, 485 Conditionally stable system block diagram, 612 frequency response design method, 339 root locus, 339 root locus extension methods, 612 stability properties for, 339–340 Constant closed loop magnitude, contours of, 383 Continuation locus, 232 Continuity relation, 54 Continuous and digital implementation, comparison of, 568

Continuous and discrete systems Simulink simulation, 489 step response comparison, 490 Continuous signal Laplace transform, 565–566 Continuous system, block diagram for, 560 Contour evaluation, 320, 321 Control canonical form, 425 block diagram, 425 derivation, 447 equations, 426 third order system, 445 transfer function to state space (tf2ss), 426 transformation to controllability matrix, 430 Control characteristic equation, 444 Control law and estimator, 478–489 finding, 443–444 for pendulum, 445–448 zero location affecting, 450–451 Control law design for full state feedback, 442–455 Ackermann’s formula, 448–452 exercise problems related to, 539, 542 finding, 443–451 pendulum example, 445 reference input, 451–455 system diagram, 444 zero location example, 450 Control responses of analog and digital implementations, 260 Control system design, 660–743 case studies, 667–739 historical perspective, 739 outline steps, 662–666 summary, 666

803

Control theory and practice, 203–205 Controllability, 432 Controllability matrix, 430 Controllable systems, 430 Controllers, 5 continuous and discrete comparison plots, 259, 355 Simulink diagram, 354, 489 dimension, 525 equations, 492, 494 polynomial form, 524 transfer function, 243–245 conv.m, 92, 459 Convolution, 90, 763 Convolution integral, 79–80 Copernicus, Nicholas, 61 Cosine signal derivative, 762 Cost function, 146, 665 Cover-up method of determining coefficients, 91 Cross over at -1, 343, 378 Crossover frequency, 335 stability margins, 334 Cruise control model, 22–24 Cruise control step response, 418–419 Cruise control transfer function using MATLAB, 98 Current source symbol and equation, 42

D D/A converters, 578 digitization, 560 Dakota autopilot, 261 Damped natural frequency, 112 Damping locus z-plane, 566 Damping ratio, 112 versus overshoot, 118 versus phase margin, 337

804 Index Damping response in digital versus continuous design, 588–589 damp.m, 588, 589 DC gain Final Value Theorem, 94 z-plane, 568 DC motor lag compensation, 363–365 lead compensation, 351–356 modeling, 48–50 reference input, 454 sketch, 47 transfer function using MATLAB, 98–99 DC motor actuators, 47 DC motor example, 184 DC Motor Position Control PI control for, 190–191 system type for, 184 DC servo example, 483–486 modified dominant second order pole locations, 489 reduced order estimator, 486–487 root locus, pole assignment, 485 DC servo compensator, redesign of, 487–489 Decade, 305 Decibel, 305 Delay contrasting methods of approximating, 273 hold operation, 560, 561 Delayed sinusoidal signal, 760 demethylation, 735 Departure angles, root locus design for, 245 Derivative, discrete control laws, 585 Describing function, 601, 619, 620 conditionally stable system, 626

for hysteresis nonlinearity, 624 for relay nonlinearity, 623 root locus extension method, 620 for saturation nonlinearity, 621 stability analysis, 625 stability analysis for hysteresis nonlinearity, 628 Design considerations, frequency response design method, 371 Design criterion for spacecraft attitude control, 343–346 Design parameters for lead-networks, 355 Design synthesis, 119 Design trade-off, 9 satellite, 461 yaw damper, 695 Desired gain, graphical calculation of, 235 Desoer, C. A., 531 Differential equations exercise problems related to, 66, 69 geometry, 415 state-variable form dynamic models, 417 Differentiation, 90, 761 Digital autopilot, 591 Digital control, 198–203, 558–592 digitization of PID, 199 discrete design, 583–590 discrete equivalents, 194, 568–577 dynamic analysis of discrete systems, 196–197, 561–568 sample-rate selection, 580–583 trapezoid rule, 194 Digital controller block diagram, 560 digital implementation, 198

space station attitude, 573–574 Tustin’s method, 570–571 Digital implementations, 257–260 lead, 373 lag, 614 PID, 192 1/s2 , 632 Digital PID, 199 Digital to analog (D/A) converters, 578 Digital versus continuous design, damping and step response in, 588–589 Digitization, 559–561 Diophantine equation, 524 Dirac, Paul, 78 Direct design with rational transfer functions, 524–527 Direct digital design. (see Discrete design) Direct transfer function formulation, 525 Direct transmission term, 416 Discrete control laws, 585 Discrete controller, 488 Discrete design, 583–590 analysis tools, 583 example, 584, 586–588 feedback properties, 585–586 Discrete equivalent of motor speed control, 201 Discrete equivalents, 200 design using, 568–577 Discrete root locus, 584–585 Discrete signals, 559 Discrete systems and continuous Simulink simulation, 489 step response comparison, 490 dynamic analysis of, 561–568 Discrete time functions to z-transform and Laplace transforms, 562 Discrete transfer function, 584

Index Disk drive servomechanism, 508–514 Disk read/write mechanism graphical illustration, 30 schematic, 31 Distinct real roots, 92 Distributed parameter systems, 38, 40 Disturbance, 4 Disturbance rejection, 1 digital control, 581–582 properties for robust servomechanism, 513 sample rate selection, 581–582 steady-state tracking by extended estimator, 518–519 Divide and conquer state space design, 415 Dominant second order poles, 456–457 Double integrator plant, 243 transfer function, 236 discrete model, 632 Double pendulum, 65 Double precision, 580 Doyle, J. C., 520, 532 Drebbel’s incubator, 10 Duality of estimation and control, 472 Dutch roll yaw damper, 691 Dynamic analysis of discrete systems, 561 Final Value Theorem, 566–568 s and z relationship, 564–566 z-transform, 561 z-transform inversion, 564 Dynamic compensation, 248 exercise problems related to, 285, 289 Dynamic models, 20–63 electric circuits, 41–45 electromechanical systems, 45–50 equations for, 63

exercise problems related to, 64 fluid flow models, 50 heat and fluid flow models, 50–60 historical perspective, 60–62 linearization, 602–603 mechanical systems, 21–41 room temperature, 51 scaling, 602–603 state-variable form differential equations, 417 Dynamic response, 75–151 amplitude and time scaling, 147–149 effects of zeros and additional poles, 120–129 exercise problems related to, 151 experimental data, obtaining models from, 140–147 historical perspective, 149–150 Laplace transforms, 75–102 perspective, 74 pole locations, effect of, 108–116 state equations, 436–442 time-domain specifications, 116–120 stability, 130–140 system modeling diagrams, 102–108 Dynamic system with saturation, 610

E E. coli, 731 E. coli genome, 732 E. coli motion, 737 Effect of zero using MATLAB, 123, 124, 125 Eigenvalue, 432 Eigenvector, 432 eig.m, 433, 435

805

Einstein, Albert, 62 Electric circuits differential equation determination, 42–43 dynamic models of, 41–45 elements, 42 equations and transfer functions, 44, 45 exercise problems related to, 69, 71 Electric power line conductor, 168 Electromagnet, 45–46 Electromechanical systems, dynamic models of, 45–50 Elementary block diagrams, 104 Emulation, 568 Emulation design, 568–577 applicability limits, 576 damping and step response example, 588–590 digital approximation methods, comparison of, 575–576 MMPZ method, 575 MPZ method, 571 space station attitude digital controller, 573–574 stages, 568 Tustin’s method, 568–569, 570–571 EPROM, 580 Equilibrium, 603, 687 Equivalent gain analysis frequency response, 619 root locus, 609 Error constants, 181 Error space approach, robust tracking, 505–515 definition, 505 design, 514–515 robust control equations, 506 Estimate error characteristic equation, 467 Estimator, 443

806 Index control law combined, 478–489 and controller mechanism, 479 extended, 515 Estimator design, 466–477 exercise problems related to, 543, 549 full order, 466–472 pole selection, 476–477 reduced order, 472–475 pendulum example, 474–475 for simple pendulum, 468–469 SRL estimator design for pendulum, 477 Estimator equations, 517 Estimator error equation, 493 Estimator modes uncontrollability, 500 Estimator pole selection, 476–478 Estimator SRL equation, 477 Evans, W. R., 13, 220, 275 Evans form of characteristic equation, 223 Evans method, 222 Exact discrete equivalent, 583 Excitation-Inhibition Model, 754 Expensive control, 464–465 Experimental data obtaining models, 140, 147 sources, 141 Exponential envelope, second-quarter system response, 114 Exponential order, Laplace transforms, 757 Exponentially decaying sinusoid, 761 Extended estimator, 515–519 motor speed system, block diagram for, 519 steady-state tracking, 518–519

system for tracking, block diagram for, 517 Extra pole, effect of, 128

F Factored zero-pole form, 97 Faraday, Michael, 61 Fast poles, 110 Feedback advantage, 176 amplifier, 41, 386 analysis, 6 design trade-off, 9 Drebbel’s incubator, 275 equations, 172 liquid-level control, 9 perspective, 170 Feedback control, 2 components, 3 Feedback law with integral control, 503 Feedback loop, 735 feedback.m, 107, 263, 588 Feedback of state, 443 Feedback output error to state estimate equation, 467 Feedback properties discrete design, 586 discrete design example, 585 Feedback scheme for autopilot design, 262 Feedback structure, 243–245 Feedback system, 104 actuator saturation, 616 drawing, 373 Feedforward of disturbances, 664 integral model example, 515 Fibonacci numbers, 593 Filament, 733 Final Value Theorem, 92–94 DC gain, 94 example, 93 incorrect use of, 93–94

stable system, 93 Finite zeros, 97 First order system, SRL for, 459 First order term, 307 Fixed point arithmetic, 581 Flagellum, 733 Flexible disk read/write mechanism, 30–32 Float valve, 9 Fluid flow models, 50 Fly-ball governor, 10 Fly-by-wire, 740 Folding z and s-plane, 566 Forced differential equations solution, 95 Forced equation solution with zero initial conditions, 95–96 Forcing function, 438 Fourier, 150 Fourth order system in modal canonical form, 428 Foxboro Company, 204 Franklin, G. F., 530 Fraser, Don, 591, 740 Free body diagram disk read/write mechanism, 31 Newton’s law, 22 suspension system, 26 Free running, 559 Frequency control system design procedure, 662 Frequency response, 13, 82, 297–317 characteristics of capacitor, 299–300 of lead compensator, 300–304 control system design procedure, 663 example, 82, 84 exercise problems related to, 389–393 lag lead compensation, 498

Index lead compensation, 349 PD compensation, 348 state space design methods, 481 Frequency response data experimental data source, 141 phase margin, 336 Frequency-response design method, 296–388 Bode’s gain-phase relationship, 341–346 closed-loop frequency response, 346–347 compensation, 347–380 control system design, 297–317 data, alternative presentation of, 382–385 exercise problems related to, 389–412 historical perspective, 386 lag compensation, 360–365 neutral stability, 317–318 Nyquist stability criterion, 319–333 perspective, 296–297 PI compensation, 360 stability margins, 334–341 time delay, 381–382 Frequency response plots, 299 LTR design, 522 Frequency response resonant peak versus phase margin, 337 Frequency shift, 90, 761 Frieden calculator, 276 Fuel-air ratio control, 742 Fuel injection, 704 Full order compensator design for DC servo, 483–486 for satellite attitude control, 480–482 Full-order estimator, 466 Fuller, 10

G Gain calculation for reference input, 452 Gain margin frequency response design method, 334 LTR, 519 magnitude and phase plot, 335 Gain phase relationship demonstration, 341 Gain selection, 501–502 Gain stabilization, 256, 371 Galileo, Galilei, 61 “Gap between Theory and Practice,” 204 Gardner and Barnes, 150 General controller in polynomial form, 524 GM, 334 Golden Nugget Airlines Problem, 292, 407 GP-B satellite, 537 Gunkel, T. L., 530 Gyroscope, 689

H Halley, Edumund, 60 Hanging crane for rotational motion, 36–37 Hard disk control system design procedure, 709, 717 Hardware characteristics, 577–580 Heat exchanger illustrated, 52 modeling equations for, 52–54 pure time delay, 529–530 root locus, 274 Heat exchanger tuning, 195–196 Heat flow, 50 equations for, 51–52 models dynamic models, 50

807

exercise problems related to, 71, 73 Heat transfer RTP system, 722 Heaviside, Oliver, 41, 149 Helicopter near hover, 283, 549 Hessenberg Matrix, 542 High frequency plant uncertainty, effect of, 372 Historical perspective, 60, 149, 203, 274, 386, 530, 590, 648, 739 Hold operation, delay due to, 560, 561 Homogeneous differential equations solution, 94–95 Hot-air balloon, 746 Hurewicz, W., 590 Huygens, Christian, 11 Hydraulic actuators, modeling, 58–60 Hydraulic piston, modeling, 55–56

I Illustrative root locus, 236 Impulse function transform, 88 impulse.m, 110, 115, 127 Impulse response, 78, 108, 113 using MATLAB, 110 Impulse signal, 77 Incompressible fluid flow, 54–60 Incubator, Drebbel, 9 Inductor symbol and equation, 42 initial.m, 469, 475 Initial Value Theorem, 769–771 Inner loop design, 697 Input filter, 5, 177 Input generator, internal model principle of, 508 Input magnitude, nonlinear example of stability, 611 Input matrix, 416 Integral, discrete control laws of, 586

808 Index Integral control, 502–519 augmented state equations, 503 block diagram, 504 description, 502–504 exercise problems related to, 554–556 feedback law, 503 internal model approach, 508 motor speed system, 503–505 polynomial solution, 527 using error space design, 514–515 structure, 503 Integral model with feedforward, 515 Integration, 90, 762 Integrator, 421 op-amp, 44–45 Integrator anti-windup, 615–618 Internal model approach, integral control using, 508 Internal model principle, 508 Internal stability, 132 Internet modeling, 741 Interrupt, 559 Inverse Laplace transforms by partial-fraction expansion, 91–92, 766–769 Inverse nonlinearity, 603 Inverse transform, 86 Inverted pendulum, 37 equations, 38 SRL estimator design, 477 example, 460–462 step response, 462 inv.m, 434, 436 Isoclines, 630

J James, H.M., 590 Joseph, P. D., 530

K Kalman, R. E., 14, 530, 640, 649 Kepler, Johannes, 61 Keynes, John Maynard, 61 Khalil, H., 620, 640 Kharitonov theorem, 140 Kirchhoff’s current law (KCL), 41 Kirchhoff’s voltage law (KVL), 41 Kochenberger, R. J., 649 Kuo, B., 50

L Lag compensation, 252–253 circuit illustration, 261 compensation characteristic, 370 DC motor, 363–365 definition, 248 design using, 254–255 design procedure, 361 frequency response design method, 360–365 state space method, 496–499 Lagrange, 648 Lamp nonlinearity, 727 Lancaster, F. W., 697 LAPACK, 433 Laplace transforms, 75–102, 436, 757–771 continuous signal, 564–566 definition, 87 differential equations solving, 96 dynamic analysis, 561 exercise problems related to, 156 homogeneous differential equations solution, 94–95 impulse function transform, 88 for problem solving, 94–96

properties, 86, 89–91, 757–766 simple discrete time functions, 562 sinusoid transform, 88 step and ramp transforms, 88 to solve problems, 94 table, 758 LaSalle, L. P., 640 Law of generators, 46 Law of motors, 45 Lead compensation, 249, 348–351 analog and digital implementations, 258 circuit illustration, 259 compensation characteristic, 370 for DC motor, 351–356 definition, 248 design procedure, 356, 664 design using, 249–254 discrete control laws, 586 frequency response, 349 maximum phase increase, 350 primary design parameters, 355 using MATLAB, 264 Lead compensator, 238 design for type 1 servomechanism system, 358–359 example, 238–239 frequency response characteristics, 300–301 state-space example, 483 temperature control system, 356–358 Lead ratio, 350 Least common multiple, extended estimator, 516 Least-squares system identification, 146 Left companion matrix, 471 Left half plane (LHP) effect of zero in, 238 Final Value Theorem, 566

Index LTR, 519 root locus extension method, 273 SRL, 458 zeros, 315 LHP. (see Left half plane (LHP)) Libler, S., 734 Ligand, 734 Limaçon, 333 Limit cycle, 613 Linear differential equations solutions, 603 standard form, 416 Linear quadratic based compensator design, 520 Linear quadratic regulator (LQR) gain and phase margins, 466 Nyquist plots, 461 regulator poles, limiting behavior of, 464–465 regulators, robustness properties, 465–466 tape drive, 463–464 Linear systems analysis using MATLAB, 97–102 ways of representing, 98 Linear time invariant systems for stability, 129–132 Linearization, 602 control system design procedure, 664 definition, 602 dynamic models, 602–603 by feedback, 602–603 by inverse nonlinearity, 608 of motion in ball levitator, 604–607 nonlinear feedback, 608 nonlinear pendulum example, 604 rapid thermal processing example, 609 by small-signal analysis, 603

water tank height and outflow, 56-57 Liquid level control, 9 Locus, real-axis parts of, 228 Logarithmic decrement, 163 loglog.m, 84, 300, 315, 521 logspace.m, 84 Long division in z-transform inversion, 562 Loop transfer recovery (LTR), 519–523 design frequency response plots, 522 design for satellite attitude control, 521–524 example, 521 Loudspeaker geometry, 45–46 modeling, 45–46 with circuit, 46–48 Low sensitivity, design criteria for, 372 LQ, 530 lqe.m, 521, 725 LQF, 530 LQG, 532 lqr.m, 463, 725 LQR. (see Linear quadratic regulator (LQR)) lsim.m, 101, 393 LTI systems, stability of, 131–132 LTR, 519–523 Luenberger, D. G., 531 Lumped parameter model, 40 Lyapunov, A. M., 12, 601 Lyapunov direct method for position feedback system, 639–640 Lyapunov equation, 637 Lyapunov first method, 638 Lyapunov function, 602, 637 Lyapunov indirect method, 638 Lyapunov stability, 636–642 definition, 636

809

exercise problems related to, 656, 659 nonlinear system, 639 second-order system, 638–639 Lyapunov stability analysis, 636 asymptotically stable, 636 first method, 636, 638 redesign of adaptive control, 640 second method, 636, 637 stable in the large, 636 Lyapunov stability theorem, 637

M M and N circles, 383 MacFarlane, A. G. J., 532 Magnetic levitation, 167, 753 Magnitude frequency response, 299 transfer function class, 307 Magnitude condition, 234 parameter value, 282 Magnitude plot gain and phase margin, 335 transfer function class, 308 margin.m, 357, 397, 521 Matched pole-zero (MPZ) method, 575 emulation design, 571 MMPZ comparison, 576 Mathematical model, 6 MathWorks, 15 MATLAB acker, 449, 472 pole location, 456 axis, 329, 332 bode, 83, 300, 315 c2d, 488 canon, 435 commands, 788–791 computing roots by, 136 conv, 92, 459

810 Index eig, 433, 435 ezplot, 137 feedback, 107, 263, 588 impulse initial, 110, 115, 127 impulse response by, 115 initial, 469, 475 inv, 434, 436 linear system analysis, 97–102 linmod, 603 linmod2, 603 loglog, 84, 300, 315 logspace, 83 lqe, 521, 725 lqr, 463 lsim, 101, 393 margin, 357, 397, 521 max, 380 nichols, 383 nyquist, 326, 329, 332 ones, 101 pade, 273 parallel, 107 place, 449, 472 plot, 33, 101 poly, 96 printsys, 98 pzmap, 109 residue, 92, 96 rlocfind, 246, 399 rlocus, 225, 398, 459 rltool, 282 roots, 441 semilogx, 84, 300, 521 semilogy, 380 series, 107, 521, 588 sqrt, 509 ss2tf, 423, 438 ss2zp, 423 step, 33, 99 step response, 24 tf, 33, 84, 107 tf2ss, 426 tf2zp, 100, 423 tzero, 440, 442 max.m, 380

Maxwell, James Clerk, 11, 62, 204, 275, 649 Mayr, O., 9 Mechanical systems dynamic models, 21–24 exercise problems related to, 66 Méchanique céleste, 149 Mello, B. A., 736 Method of computed torque, 603 Microphone, 70 Microprocessors for control applications, 580 Minimum phase systems and Bode plot, 316 MIT rule, 641, 649 Mixed control system versus pure discrete equivalent, 584 MMPZ method emulation design, 575 MPZ comparison, 576 Modal canonical form block diagram, 427 fourth-order system block diagram, 428 state equations, 427–428 Modal form transformation, 432, 433–434 Model, definition, 20 Modern control, 14 Modern control design, 415 Modes of the system, 97 Modified dominant second order pole locations with DC servo redesign, 489 Modified matched pole-zero (MMPZ) method emulation design, 575 MPZ comparison, 576 Modified PD control, 238–239 Moler, Cleve, 277, 532 Monic, 222 Monic polynomials, 524 Motion equation development for rigid bodies, 40 Motor

AC, 49 DC, 47 position control, root locus for, 223–224 Motor speed, PID control of, 189 Motor speed system extended estimator, 519 integral control, 503–504 MPZ method emulation design, 571–575 MMPZ comparison, 576 MRFM, 740 Multiple crossover frequency system, Nyquist plot for, 340–341 Multiplication by time, 91, 765 Multipoint injection, 704 Munro, N., 532

N Napoleon, 149 Natural frequency, 438 z-plane, 566 Natural mode, 438 Natural responses, 108 stability, 114 N circles, 383 Negative feedback, 104 Negative locus, 226 definition, 268 plotting rules, 266–269 sketching, 269 Negative root locus. (see Negative locus) Neutral stable system, 132 Neutrally stable, 132 frequency response design method, 317–319 exercise problems related to, 394 Newton, Isaac, 60 Newton’s law, 60 rotational motion, 27 translational motion, 21

Index Nichols, N.B., 590 Nichols chart, 382 contours of constant closed loop magnitude, 383 for PID example, 383–385 nichols.m, 383 Noncollocated case, root locus for, 245 Noncollocated sensor, 244, 246 Nonlinear differential equations, 603 Nonlinear integrator, 656 Nonlinear radiation RTP system, 722 Nonlinear sensor, 704 Nonlinear systems, 599, 600 analysis and design based on stability, 629–648 equivalent gain analysis using frequency response, 619–629 equivalent gain analysis using root locus, 609–619 historical perspective, 648–649 linearization, 602–609 Lyapunov stability, 639 root locus extension methods, 618–625 Nonlinearity complications, 705 Nonminimum phase systems Bode plot, 315 frequency response, 315 LTR, 520 response of, 127 Nonminimum phase zero, 127 Normal form, 443 state space design, 414 Normal modes, 427 Notch compensation definition, 249 design using, 255–257 root locus extension method, 614

Notch filter block diagram, 615 Nyquist, H., 13, 386 Nyquist and Shannon sampling theorem, 578 Nyquist frequency, 566 nyquist.m, 326, 329, 332 Nyquist plot, 323 characteristics, 332–333 defining gain and phase margin, 334 evaluation, 322 LQR design, 461 multiple crossover frequency system, 340 open loop unstable system, 330–332 plotting procedure, 324 pole locations, 460 second order system, 324–326 third order system, 327–329 using MATLAB, 329 vector margin, 338 Nyquist stability criterion, 319–332 exercise problems related to, 394–396

O Observability, 471 Observability matrix, 471 Observer, 443 Observer canonical form, 431, 469–471 block diagram, 470 DC servo, 485 equation, 470 illustrated, 431 third order system, 470 oil-mass resonance, 60 One-sided Laplace transforms, 87, 757 ones.m, 101

811

Op-amp as integrator, 44–45 schematic symbol, 43 simplified circuit, 43 summer, 44 Open loop autopilot design, 262 Open loop control, 7 system, 170, 172 Open loop estimator, 467 Open loop transfer function, 318, 630 Open loop unstable system, Nyquist plot for, 330–332 Optimal control, 457 Optimal design, 664 Optimal estimation, 477 Oscillator, time scaling, 424 Oscillatory systems block example, 613 with saturation, 613 Oscillatory time response, 115 Output equation, 437 state description, 428 Output matrix, 416 Output response, analog and digital implementations, 259 Overshoot definition, 116 plot, 122 time domain specification, 116–118 versus damping ratio, 118

P Padé approximant, 273 pade.m, 273 Paper machine example, 17 parallel.m, 107 Parameter, consideration of two, 270–272 Parameter range versus stability, 135–139 Parameter selection, 669

812 Index Parameter value selection, 234, 236 Parseval’s theorem, 645, 765 Partial fraction expansion distinct complex roots, 766–768 frequency response, 298 inverse Laplace transform, 91–92, 766–769 repeated real roots, 768–769 Partial state control canonical form, 446 Peak amplitude, 309 Peak time, 116 Pendulum example control law, 445 estimator design, 468–469 linear and nonlinear response, 604 nonlinear equations, 32 reduced order estimators, 474–475 rotational motion, 32 SRL estimator design, 477 rotational motion step input, 34 Performance bound function example, 374–377 plot, 375 Perturbations, 506 Phase, 83 frequency response, 299 Phase condition, 226 Phase lag between output and input, 85 versus time delay, 381 Phase margin closed loop bandwidth, 347 frequency response data, 336 frequency response design method, 334 LTR, 519 magnitude and phase plot, 335 versus damping ratio, 337 versus frequency response resonant peak, 337

versus transient response overshoot, 337 Phase plane, 630 state space design, 415 Phase plot gain and phase margin, 335 transfer function class, 308 Phase stabilization, 256, 371 Phase-plane plot with saturation, 632 Photolithography, 754 Phugoid mode, 697 PID control, 13 Piper Dakota autopilot, 262 Pitch angle, 685 place.m, 451, 472, 489 Plant, 5 estimator connected to, 469 Plant changes, 666 Plant evaluation/modification, 664 Plant inversion, 520 Plant open loop pole, root locus with, 224–225 Plant uncertainty example, 377 plot, 376 plot.m, 33, 101 PM, 334 PM to damping ratio rule, 337 Poincaré, 648 Polar plot, 323 Pole Bode plot, 310–311 compensator design, 549 correlation, 115 definition, 97 inverse Laplace transforms, 91 partial fraction expansion, 768 rational transfer function, 96 response character indication, 96 of the system, 97 and zeros, finding using MATLAB, 436 Pole assignment

root locus DC servo, 485 SRL, 488 Pole location/placement, 108, 115 Ackermann’s formula, 448 corresponding impulse responses, 114 exercise problems related to, 158 polynomial transfer functions, 525–526 selection, 455 dominant second order poles, 455–457 example, 456–457, 459–464 exercise problems related to, 542, 543 methods, 455, 466 SRL, 457–465 Pole selection SRL, 490–491 Pole-zero cancellation, 500 Pole-zero patterns, effects of, 128–129 poly.m, 96 Polynomial solution, integral control to, 527 Polynomial transfer functions pole placement for, 525–526 reduced-order design for, 526–527 Pontryagin, L. S., 14 Popov, 649 Position error constant, 179, 316 Position feedback system, Lyapunov direct method for, 639–640 Positive feedback, 104 Positive locus, 223, 226 plotting rules, 228–233 Positive root locus. (see Positive locus) Prefilter, 727 printsys.m, 98

Index Problem solving, using Laplace transforms, 94 Process, 4 Process noise for estimator pole selection, 476 Process reaction curve, 193 Programmable read only memory (EPROM), 580 Prokaryotic cell, 731 Proportional derivative (PD)compensation, 348 compensation characteristic, 370 frequency response, 348 Proportional feedback control (P), 187 discrete control laws, 585 Proportional integral (PI) anti-windup compensation, 618 compensation, 360 control system, 137 compensation characteristic, 370 Proportional plus integral control (PI), 187–188 Proportional-integral-derivative (PID), 171 controller, 186 frequency response design method, 365–371 Nichols chart as example for, 383–384 spacecraft attitude control, 365–371 state-space design, 664 Proportional-integral-derivative (PID) control, 188–189 digital form, 192 Prototype testing, 665 Proximate time-optimal system (PTOS), 634, 635 Pseudorandom binary signal (PRBS), 142 Pseudorandom-noise data, 141, 142 Pulse, digitization, 559

Pure discrete equivalent versus mixed control system, 584 Pure time delay design for system, 527–530 exercise problems related to, 556 heat exchanger, 529–530 Pyrometer, 720 pzmap.m, 109

Q QR algorithm, 433 Quality factor, 112 Quantized signals, 198 Quarter car-model, 25 Quarter decay ratio, 194

R Radiation Laboratory, 204 Ragazzini, J. R., 41, 590 RAM, 580 Ramp response of first order system, 763 Ramp signal, robust tracking, 506 Random access memory (RAM), 580 Random walk, 735 Rapid thermal processing (RTP) system control system design procedure, 717 laboratory model, 722 linear model, 723 Rational transfer functions, direct design with, 523–527 Reaction jets, 668 Reaction wheels, 668 Read-only memory (ROM), 580 Receptor, 734, 735 Reduced order compensator design for satellite attitude control, 482–483

813

Reduced order compensator transfer function, 480 Reduced order design for polynomial transfer function model, 526–527 Reduced order estimator, 472–475 DC servo redesign example, 486–487 for pendulum, 474–475 structure, 474 Reference input alternative structures, 455 estimator, 491 gain selection, 501 general structure, 492–501 example, 452–453 full state feedback, 451–455 gain equation, 452 selection methods, 493 type 1 system, 454 Reference spectrum plot, 373 Reference tracking, 171 Region of convergence, 757 Regulation, 171, 174–175 Regulation with disturbance inputs for system type, 183 Regulator, 16 compensator design, 549 Relay, 610 Repeated poles, 768 Repeated real roots, partial fraction expansion, 768–769 Repellent, 733–734 Reset control, 616 Reset windup, 616 residue.m, 92, 96 Resistor symbol and equation, 42 Resonant frequency, 383 Resonant peak, 302 Response by convolution, 75–79 to sinusoid, 82 versus pole locations and real roots, 109–111

814 Index Right half-plane (RHP), 112 Bode plot, 332 compensator transfer function, 480 LTR, 520 Lyapunov stability, 638 root locus extension method, 273 SRL, 458 zeros, 123 example, 127 nonminimum-phase systems, 315 Rigid bodies, motion equation development for, 40–41 Ring-lasers gyroscope, 690 Rise time, 116 rlocfind.m, 246, 399 rlocus.m, 225, 398, 459 RLTOOL, 282 RMS value, 523 Robust, 506 Robust control definition, 506 error space equations in, 506 sinusoid, 508–514 Robustness, 2 Robust properties, 181, 506 LQR regulators, 465 system type, 181 Robust servomechanism closed-loop frequency response, 513 disturbance rejection properties, 513 Simulink block diagram, 511 tracking properties, 512 Robust tracking, 502–519 error space approach, 505–515 exercise problems related to, 554–556 Robustness constraints, 378–379 Rolling mill, 156 Roll mode, 691 ROM, 580

Room temperature, dynamic model for, 51 Root locus, 14, 220 0 degree locus. (see Negative locus) 180 degree locus. (see Positive locus) autopilot design, 263 closed loop poles, 271 combine control and estimator, 482 compensation, 614 complex multiple roots, 247 conditionally stable system, 339 DC servo pole assignment, 485 reduced order estimator, 486 definition, 226 design example, 260–266 exercise problems related to, 279, 282, 289–293 feedback system, 221, 226 guidelines for determining, 226 illustrative, 236–248 lead design, 251 lead lag, 257 method of Evans, 222 motor position control, 223 noncollocated case, 245–247 notch compensations, 257 plant open-loop pole, 224 plotting rules, 233 reduced order controller, 483 rules application example, 245 rules for plotting a positive, 228–233 for satellite attitude control, 236–238 satellite control collocated flexibility, 243 lead compensator, 238–239

small value for pole, 239–240 transition value for pole, 240–243 sketching guidelines, 226 exercise problems related to, 279 SRL pole assignment, 488 stability examples, 318 time delay, 272 heat exchanger, 274 using two parameters in succession, 270–272 Root locus design method, 220, 278 basic feedback system, 221–226 design example using, 260–266 dynamic compensation, design using, 248–260 exercise problems related to, 279–295 extensions of, 266–274 historical perspective, 274–276 guidelines for determining root locus, 226–236 illustrative root loci, 236–248 perspective, 220 Root locus forms, 223 Root locus method extensions, 270–272 exercise problems related to, 293 of Evans, 222 Root mean square (RMS) value, 523 roots.m, 136, 441 Rosenbrock, H. H., 532 Rotational and translation hanging crane, 36 Rotational motion Newton’s law, 27 pendulum, 32

Index satellite attitude control model, 28–30 Rotor, free body diagram for, 48 Routh, E. J., 12, 204, 275 Routh array, 133–134 Routh’s stability criterion, 132–134 Routh’s test, 134–135, 139 RTP laboratory model, 722 RTP linear model, 723 RTP system, 717–730 Rudder, 685 Run, E. coli, 733 Run-to-run control, 753

S Saberi, A., 520 Safonov, M., 532 Sample period, digitization, 559 Sample rate digitization, 559 lower limit, 580 Sample rate selection digital control, 580–583 anti-alias prefilter, 582–583 asynchronous sampling, 583 disturbance rejection, 581–582 tracking effectiveness, 581 digitization, 560 Sampled data system, 559 Sampled signals, 198 Sampling theorem of Nyquist and Shannon, 578 Sandberg, I. W., 647 Sastry, S., 640 Satellite attitude control, 191 application, 667–684 example

full order compensator design for, 480–482 LTR, 521 reduced order compensator design, 482–483 rotational motion, 28–30 SRL, 459–460 state-variable form, 417–418 Satellite control root locus collocated flexibility, 243 lead compensator, 238–239 small value for pole, 239–240 transition value for pole, 240 schematic, 29 Satellite transfer function using MATLAB, 101–102 Satellite with flexible appendages, 313–315 Saturation dynamic system, 610 oscillatory systems, nonlinear example, 613 Scaling ball levitator, 147–148 dynamic models, 602–603 time, 89 Schmitt trigger circuit, 624 Schmitz, E., 40 Second order equations, external signals, 516 Second order servomechanism, 496–499 Second order step responses of transfer functions, 124 Second order system block diagram, 105 Lyapunov stability, 638–639 Nyquist plot, 324–326 responses, 113 exponential envelope, 114 step response plots, 122

815

Second order term, transfer function class, 308 Segway, 38 Semiconductor wafer manufacturing, 717 semilogx.m, 84, 300, 521 semilogy.m, 380 Sensitivity, 171, 175–178 Sensitivity function, 373 for antenna, 380 design limitations, 377–381 plot and computation, 380–381 specifications, 373–381 exercise problems related to, 410 Sensor, 5 control system design procedure, 662 Sensor noise, 371 estimator pole selection, 476 Sensors for Control, 204 series.m, 107, 521, 588 Servo motor, torque-speed curves, 49 Servo motor. (see also DC servo) Servo speed control, SRL for, 459 Servo with tachometer feedback, system type for, 182 Servomechanisms, 13 increasing velocity constant, 496–499 structure, block diagram of, 271 system, lead compensator design for, 358–359 Set point, 4 Settling time definition, 116 time domain specification, 118–120 Shift in frequency, 761 Short pulse, 77 Short-period mode, 697

816 Index Sifting property, 78 Signal decay, rate of, 110 Simon, H. A., 531 Simple design criterion for spacecraft attitude control, 343–346 Simulation block diagram transfer function, 422–424 control system design procedure, 665 Simulink, 737 block diagram for LTR, 523 block diagram for robust servomechanism, 511 for nonlinear motion, 35–36 Simulink nonlinear simulation definition, 709 RTP system, 727 Simulink simulation for continuous and discrete systems, 489 Single point, 704 Sinusoid robust control, 508–514 with frequency, 760–761 Sinusoid of frequency, compensator structure, 509 Sinusoid transform, 88–89 Sinusoidal signal, 758 time integral, 762 time product, 765–766 Sizing the actuator, 619 Slow poles, 110 Small signal linearization, 603–604 Smith compensator, 528 Smith regulator for time delay, 528 Solutions to end-of-chapter questions, 772 Space station attitude digital controller, 573–575 Space station digital controller, direct discrete design of, 586–587 Spacecraft attitude control

PID compensation design for, 366–371 simple design criterion for, 343–346 Specific heat, 51 Spectral analyzers, 301 Sperry, L., 739 Spiral mode, 691 Spirule, 275 S-plane complex poles, 112 RHP contour, 322 specification transformation, 120–121 time domain specification, 119 and z-plane relationship, 565–566 sqrt.m, 509 SRL. (see Symmetric root locus (SRL)) ss.m, 101, 435 ss2tf.m, 423, 428 ss2zp.m, 423 Stability, 1, 108, 129–140, 173 exercise problems related to, 166, 169 feedback system, 135 input magnitude, nonlinear response, 611 linear time invariant systems, 129–132 Lyapunov, 636 natural responses, 114 necessary condition for, 133 Routh’s criterion, 232 of system, 272 system definition and root locus, 318 versus parameter range, 135–137 versus two parameter ranges, 137, 139 Stability analysis, 12 Stability augmentation, 685 Stability condition, frequency response, 319

Stability margins frequency response design method, 333, 341 exercise problems related to, 396–401 Stability properties for conditionally stable system, 339–340 Stability robustness, 374 Stabilization amplitude, 371 phase, 371 Stable compensator, 485 Stable minimum phase system, 341 Stable system block diagram, 612 definition, 130 Staircase algorithm, 542 Star tracker, 668 State description equation, 436 thermal system transfer function, 437–438 zeros for thermal systems from, 439 State equations block diagrams, 425–436 canonical forms, 425-436 dynamic responses from, 436–437 examples, 427–428, 433–436, 440–442 exercise problems related to, 536–539 State estimate equation, feedback output error, 467 State of the system, 416 State space and frequency response design methods, 481 State space control design, 414 State space design, 413–536 advantages, 414–416 block diagrams and, 421–424 compensator design, 478–489

Index control-law design for full-state feedback, 442–455 estimator design, 466–477 exercise problems related to, 536–557 gain selection, 501–502 historical perspective, 530–533 integral control and robust tracking, 502–519 loop transfer recovery (LTR), 519–524 Lyapunov stability, 636–642 perspective, 413 pole location selection for good design, 455–466 pure time delay, 527–530 rational transfer functions, 523–527 reference input with estimator, 491–502 state equation analysis, 425–442 system description in, 416–421 time and amplitude scaling in, 424 State space design elements, 443 state-variable form differential equations dynamic models, 417 example satellite attitude control, 417 loudspeaker with circuit in, 420–421 modeling DC motor in, 420–421 State space method, lag compensation by, 496–499 State space pole placement method, 483 State space to transfer function (ss2tf) MATLAB function, 440

State transformation, 429 Steady-state error, 315–317 command inputs and disturbances, 371 determination example, 317 Steady state information, stochastic, 141 Steady state tracking and disturbance rejection, 518–519 Steam engine, 10 Stein, G., 532 Step and ramp transforms, 88 step.m, 33, 99, 418 Step response, 113, 419 digital versus continuous design, 588–589 first-order system, 119 lag compensation, 499 lead, lag, and notch compensations, 258 MATLAB, 24 standard second-order system, 117 Step response comparison of continuous and discrete systems, 259, 490 Step response plots, second-quarter systems, 122 Stick on a cart, 37, 542 Stochastic steady state information, 141 Structurally stable, 506 Successive loop closure, 270–272 Superposition, 89, 758 example, 76 principle of, 75–76 Superposition integral, 79 Suspension model, two-mass system, 25–27 Suspension system, free body diagram for, 26 Sylvester matrix, 525 Symmetric root locus (SRL), 457–466

817

DC servo compensator redesign example, 487–489 equation, 458 estimator pole selection, 477 estimator design pendulum example, 477 first order system, 459 for inverted pendulum, 460–462 pole assignment, 487 pole selection, 489 for satellite attitude control, 459 for servo speed control, 459 System cruise control model, 21–24 definition, 304 System complexity versus system response, 695 System identification, 20, 146 System matrix, 416 System modeling diagrams, 102 exercise problems related to, 156, 159 System response versus system complexity, 695 System stability, 130 System type, 178 definition, 178 reference tracking, 183 for regulation and disturbance rejection, 183–186 robust property, 181 for speed control, 181 for tracking, 179–181 using integral control, 181 Systems biology, 731

T Tape drive, 154 analysis of state equations of, 440–442 example, 434–436

818 Index LQR design, 463–464 state equation analysis, 440–442 Tape servomotor design, step responses of, 457 Taylor Instrument Company, 204 Temperature control system lag compensator design for, 362–363 lead compensator for, 356–358 Temperature trajectory, 727 Temperature uniformity, 724 Tension plots for tape servomotor step responses, 457 Tesla, N., 49 tf.m, 33, 84, 107 tf2ss.m, 426 tf2zp.m, 100, 423 Thermal capacity, 51 Thermal conductivity, 51 Thermal resistance, 51 Thermal system transfer function from state description, 437–438 Thermal system transformation from control to modal form, 433–434 Thermal systems from state description, zeros for, 439–440 Thermostat, 5 Third order system analog computer, 422 block diagram, 423 control canonical form, 445 lead compensator design, 358–359 Nyquist plot, 327–329 observer canonical form, 470 step responses, 129 Thomson, W. T., 40 Three-term controller, 186 Throttle body injection, 704 Time constant, 108

Time delay, 89, 272–275, 760 frequency response design method, 381 exercise problems related to, 409 heat exchanger, 529–530 magnitude, 381 phase, 381 root locus, 272 heat exchanger, 274 Smith regulator, 528 transfer function, 527 Time delay system, 381 Time domain specifications, 116–121 exercise problems related to, 159, 163 Time integral of sinusoidal signal, 762–763 Time invariance, 76–77 Time multiplication, 765 Time product Laplace transform, 90 property, 764 sinusoidal signal example, 765 Time response, 663 Time scaling, 89, 148, 760 oscillator, 424 Time sequences with z-plane, 567 Time-to-double, 166 Torque, 48 Torque-speed curves for servo motor, 49 Torricelli, 648 Tou, J. T., 530 Tracking, 171, 172, 174 Tracking control, 16 Tracking effectiveness, 581 Tracking error estimator description, 493 example, 494 Tracking properties for robust servomechanism, 512 Tracking systems, 1 Trankle, T., 291

Transconduction pathway, 734 Transfer function, 80–82 block diagram, 106 Bode form, 306 cancellations, 97 classes of terms, 306 for closed-loop system, 500 example, 81–82 linear systems, 85, 87 of simple system using MATLAB, 107–108 state equations, 437, 446 time delay, 527 using MATLAB, 107 Transfer function poles from state equation, 438 Transfer function to state space (tf2ss) control canonical forms, 426 Transfer function zeros from state equations, 439 Transformation of state equations, 429 Transformations using MATLAB, 100–101 Transient response data, 141 DC motor, 100 definition, 86 experimental data source, 141 motor speed system, 505 overshoot versus phase margin, 337 Translational motion, Newton’s law for, 21 Transmission zeros (tzero), 442 Transportation lag. (see Pure time delay) Transpose, 416 Trapezoidal integration, 569 Trapezoid rule, 200, 353 Trim command loop, 265 Truxal’s formula, 185, 186, 495 Tumble, E. coli, 733

Index Tumbling frequency, 735 Tungsten halogen lamp, 720 Tustin’s method, 569 digital controller, 570–571 emulation design, 568 MPZ and MMPZ comparison, 576 Two-mass system, suspension model, 25–27 Two-sided Laplace transforms, 87, 757 Type k, definition of, 182 Type one servomechanism system, lead compensator design for, 358–359 Typical plant uncertainty, 377 tzero.m, 440, 442

U Ultimate gain, 194 Ultimate period, 194 Ultimate sensitivity method, 194 Uncontrollability of estimator modes, 500 Uncontrollable systems, 450 Undamped natural frequency, 112 Undamped oscillator, Ackermann’s formula for, 448–450 Unilateral Laplace transforms, 87, 757 Unit step function, 80 Unit-step response, 419 Unity feedback structure, 243 Unity feedback system, 104 drawing, 373 Unstable, impulse response, 108 Unstable system, 130 Upper companion form, 447 USCG Cutter Tampa, 291

V Van der Pol, B., 87 Van der Pol’s equation, 659 Vector margin, 379 Velocity constant, 180 Velocity error coefficient, 316–317 Vidyasagar, M., 640 Voice coil, 45–46 Voltage source symbol and equation, 42

W Washout, 691 Water tank height, equation for describing, 54 Water tank height and outflow, linearization of, 56–57 Watt, James, 204 Watt’s flyball governor, 10, 11, 649 Weak controllability, 450 Weighting function in Bode gain phase theorem, 343 Wiener, N., 13 Widnall, Bill, 591, 740 Winnie Mae, 739 Woodson, H. H., 596 Wright brothers, 739

Y Yakubovich, 649 Yaw damper, 689–696 Yi, T.-M., 734

Z Zadeh, L., 531 Zames, G., 532, 647 Zero, 108 Bode plot, 310–311, 313–315 definition, 97

819

effect of, 120–128 LHP, 238 exercise problems related to, 163, 166 inverse Laplace transforms, 91 LTR, 520 rational transfer function, 96–97 RHP, 124–126, 315 root locus extension method, 274 using MATLAB to find, 434–436 Zero assignment estimator description, 493 example, 496 increasing velocity constant, servomechanism, 496–499 Zero degree locus. (see Negative locus) Zero location and control law, 450 Zero order hold (ZOH), 560 digitization, 561 sample rate selection, 581 Ziegler-Nichols Tuning of PID, 192–195 Zirconia sensor, 703 z-plane natural frequency and damping locus, 566 and s-plane characteristics, 565 digital control relationship, 564–566 relationship, 565–566 time sequences, 567 z-transform dynamic analysis, 561–562 inversion, 562, 564 tables, 563 simple discrete time functions, 562

This page intentionally left blank

Table of Laplace Transforms Number

F (s)

f (t), t ≥ 0

1

1 1 s 1 s2 2! s3 3! s4 m! sm+1 1 (s + a) 1 (s + a)2 1 (s + a)3 1 (s + a)m a s(s + a) a 2 s (s + a) b−a (s + a)(s + b) s (s + a)2 a2 s(s + a)2 (b − a)s (s + a)(s + b) a (s2 + a2 ) s (s2 + a2 ) s+a (s + a)2 + b2 b (s + a)2 + b2 a 2 + b2  s (s + a)2 + b2

δ(t)

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

1(t) t t2 t3 tm e−at te−at 1 2 −at t e 2! 1 t m−1 e−at (m − 1)! 1 − e−at 1 (at − 1 + e−at ) a e−at − e−bt (1 − at)e−at 1 − e−at (1 + at) be−at − ae−at sin at cos at e−at cos bt e−at sin bt
 a 1 − e−at cos bt + sin bt b

Chronological History of Feedback Control Automotive stability augmentation systems Farm tractor auto-steering via GPS GPS Unmanned aircraft High precision disk drive control Computer-aided control design Internal model control Feedback control of automotive engines Aircraft auto-landing Apollo digital autopilot Aircraft stability augmentation LQG design Inertial navigation Microprocessor Maximum principle Dynamic programming Numerical optimization Optimal filtering Sampled data systems Root locus Nyquist stability Frequency-response tools Feedback amplifier Autopilot Stability analysis of governor Routh stability Fly-ball governor Incubator

1600s

1700s

1800s

1910

1920

1930

1940

1950

1960

1970

1980

1990

2000

MATLAB Function (.m file)

Description

Page (s)

acker bode c2d canon conv damp eig feedback impulse initial inv logspace lqe lqr lsim margin nyquist pade parallel place plot poly pzmap residue rlocfind rlocus rltool roots series sqrt ss2tf ss2zp ss ssdata step tf2ss tf2zp tf tzero

Ackermann’s formula for pole placement Bode frequency response Continuous-to-discrete conversion State-space canonical forms Polynomial multiplication Damping and natural frequency Eigenvalues and eigenvectors Feedback connection of two systems Impulse response Initial condition response Matrix inverse Logarithmically spaced frequency points Linear Quadratic Estimator design Linear Quadratic Regulator design Linear system simulation Gain and phase margins Nyquist plot Padé approximation for time delay Parallel connection of two systems Pole placement Plot function Form polynomial from its roots Pole–zero map Residues in partial fraction expansion Find root-locus gain Root locus Interactive root-locus tool Roots of a polynomial Series connection of two systems Square root State space to transfer function State space to pole–zero conversion Conversion to state space Create a state-space model Step response Transfer function to state space Transfer function to pole–zero conversion Conversion to transfer function Transmission zeros

449, 456, 469, 472 84, 300, 315, 380, 522 203, 353, 488, 499, 570 435 92, 459, 461, 767 588–589 432, 433, 439 262–263 111, 115, 127 469, 475 434, 436 84 521 463, 464, 725 101, 102 357, 398, 522 326, 329, 333 273 107 451, 472, 484, 490, 510 33, 99, 101 96 109 92, 96, 767 246, 399 225, 399, 459, 461 282 136, 239, 441, 451 107, 522, 588–889 510 438, 440, 441 423 419, 435, 440, 442 435 24, 137, 138, 142, 285 426 98, 100 24, 84, 107, 111, 127 459, 468, 470

Design Aids Closed Loop y/r

tp

Mp ⫾1%

r

1 0.9 e R

⫹

⌺

KG(s)

⫺ 0.1 t

tr ts

Im (s)

u ⫽ sin⫺1z 20 db

Amplitude ratio, T (s)

10

Resonant peak, M r

vn

0

1 0.7

⫺3 v (rad/sec)

0.1

⫺20

Re (s) s

Bandwidth, v BW

Open Loop

Design Relations

兩G兩

ts =

σ = ζ ωn

W1 1

vc v1

GM

v W2⫺1

⬔G

⫺180⬚

4.6 σ

PM v

tr =

1.8 ωn

 ωd = ωn 1 − ζ 2

1 , K0 = |G(jω)|ω=0 1 + K0 1 |E| < , ω < ω1 1 + W1 ωBW = ωc for PM = 90◦ ωBW = 2ωc for PM = 45◦ 1 Mr ∼ = 2 sin(PM/2) Mp = 5%, ζ = 0.7 Mp = 15%, ζ = 0.5 Mp = 35%, ζ = 0.3 PM ζ ∼ for PM < 70◦ = 100 ess =

vd

Y